Compare commits

..

5 Commits

286 changed files with 857341 additions and 222762 deletions
+2 -4
View File
@@ -29,10 +29,8 @@ log/
!05_software/real/sim2real/vendored/odin1_imu/build/
!05_software/real/sim2real/vendored/odin1_imu/build/libodin1_imu_bridge.so
!05_software/real/sim2real/vendored/odin1_imu/lib/*.a
# Required vendored Odin SDK libraries in the final ROS 2 deployment
!05_software/real/sim2real_ros2/src/odin_ros_driver/lib/liblydHostApi_arm.a
!05_software/real/sim2real_ros2/src/odin_ros_driver/lib/liblydHostApi_amd.a
!05_software/real/sim2real_v2/vendored/odin1_imu/lib/*.a
!05_software/real/sim2real_ros2_v2/src/odin_ros_driver/lib/*.a
# Training outputs
logs/
+47 -11
View File
@@ -14,7 +14,53 @@
| `v0.7.0` | MuJoCo 工具 | 姿态优化、IK 扫描、动力学、MPC 和 GUI 调试工具 |
| `v0.8.0` | 后期 Sim2Sim | ONNX 回放、IK/路线检查工具和比赛最终 Rough 策略 |
| `v0.8.1` | 导航打点工具 | 地图/航点编辑、路线迭代和抽样 PCD 补充包 |
| `v0.9.0` | 最终比赛部署 | ROS 2/C++ 真机闭环、Odin、CAN、导航与屏幕 UI |
| `v0.9.0` | Python Sim2Real v2 | 反馈新鲜度、Odin odom 诊断、Web 调试和安全监控增强 |
| `v0.10.0` | ROS 2/C++ 初版 | 50 Hz C++ 推理、200 Hz CAN 热路径和 ROS 2 系统集成 |
| `v0.11.0` | ROS 2 导航原型 | 简单导航、PCD 交互定位、任务点和 Web 导航调试 |
| `v0.11.1` | Odin 与站姿调参 | 完整 Odin 驱动、TensorRT、多策略切换和调参站姿 |
| `v0.12.0` | 里程计导航联调 | 纯里程计 fallback、A_min 路线、TF 冲突保护和 model_9600 |
> 原先临时归档为 `v0.9.0` 的最终 ROS 2/C++ 比赛部署已保存在 `backup/final-ros2-v0.9.0` 分支和 `backup-v0.9.0-ros2-final` 标签中,重排完成后将正式归入 `v1.0.0`。
## `v0.9.0` 的 Python Sim2Real v2
- 归档 `real/sim2real_v2` 真机部署版本,保持 `53D -> 16D` 策略观测和动作契约。
- 增加电机反馈新鲜度判断、Odin odom 诊断、命令限加速度平滑和 Web 运行时诊断。
- 保留 Python 策略运行时、ONNX/PT 模型、MJCF、Odin 接口、Web 工具和安全保护链路。
- 排除运行日志、测试日志、临时 XML 和开发交接草稿;后续 ROS 2/C++ 版本另行归档。
## `v0.10.0` 的 ROS 2/C++ Sim2Real 初版
- 归档 `real/sim2real_ros2`,将 Python 部署契约迁移到 ROS 2 Humble 与 C++ 运行时。
- 保留 53D 观测、16D 动作、50 Hz 策略循环和 200 Hz SocketCAN 电机热路径。
- 增加消息接口、硬件桥、策略运行时、命令仲裁、Nav2 配置、Docker 和 Windows Web 调试工具。
- 原始快照中的 `src/odin_ros_driver` 为空目录,因此本版本仍需外部 Odin 驱动,不能宣称传感器依赖已自包含。
- 保留原始候选 ONNX 文件以记录初版部署试验;排除计划、任务和 walkthrough 草稿。
## `v0.11.0` 的 ROS 2 Sim2Real v2
- 归档 `real/sim2real_ros2_v2`,保持 `v0.10.0` 的 ROS 2/C++ 控制契约。
- 增加 `simple_nav_node.py`、PCD 点击工具、任务点/任务序列配置和 Web 导航控制入口。
- 默认命令源从遥控切换为 `NAV`,加入简单导航状态、PCD 位姿和地图显示链路。
- 原始 `map1.pcd``map6.pcd` 分别约 49.05 MiB、44.24 MiB,归档时确定性抽样到 10 MB 以下并记录哈希。
- 原始快照中的 Odin 驱动仍为空目录;排除设备运行日志和开发草稿。
## `v0.11.1` 的 Odin、TensorRT 与站姿调参
- 归档 `real/sim2real_ros2_v2(z=0.380 hip=0.670 knee=-1.390)`,在同一 `sim2real_ros2_v2` 目录中记录真实差异。
- 首次随部署工程保留完整 Odin ROS 驱动、Apache-2.0 许可证、设备标定参数和预编译 SDK 静态库。
- 策略运行时增加 TensorRT、ONNX 回退、Rough/Crawl 模式切换、事件日志和更完整的电机失效诊断。
- 默认 Rough 策略为 `NEWmodel_1900`,默认站姿为髋俯仰 `0.670`、膝关节 `-1.390`Crawl 配置使用 IK 后端。
- `map_b.pcd` 从 1,080,047 点确定性抽样为 270,012 点,并保留原始和抽样哈希。
- 排除嵌套 Git、Odin 运行日志、缓存、开发草稿和未被配置引用的候选策略。
## `v0.12.0` 的里程计导航联调
- 归档 `real/sim2real_ros2_v2(odom)`,继续沿用 `sim2real_ros2_v2` 目录的线性演进。
- 将 Rough 策略切换为 `model_9600`,默认站姿恢复为髋俯仰 `0.550`、膝关节 `-1.125`
- Odin `custom_map_mode` 固定为纯里程计,加入 odom 新鲜度、外部 map/odom TF 冲突和任务结束交接保护。
- 增加 A_min 路线、PCD 地图编辑工具和三份抽样点云;原始大 PCD 不直接进入 Git。
- 保留完整 Odin 驱动、标定参数和 SDK 静态库;未找到的 `map_a.bin` 仍不伪造,重定位闭环不在本 Tag 声称已复现。
## `v0.4.0` 的模型变化
@@ -72,13 +118,3 @@
- 补充 `1hao.xml``2hao.xml``A_C.xml`,并为 `1B_FF.json` 补齐其引用的 `B_C.xml`
- 将两份约 915 MiB 的原始 ASCII PCD 确定性抽样为各小于 10 MB 的预览点云;抽样参数、点数和哈希记录在工具 README。
- 训练代码、MJCF、比赛策略和历史依赖锁保持 `v0.8.0` 状态不变。
## `v0.9.0` 的最终比赛部署
- 归档比赛得分 1050 所对应的 `last_not_slalom_1050` ROS 2 工作区;1050 是成绩,不是策略编号。
- 保留 53D→16D C++ 策略运行时、200 Hz CAN 硬件桥、命令仲裁、安全监控和统一启动包。
- 保留 Rough `model_6800`、Wall `model_84` 的 ONNX 与比赛 TensorRT engineCrawl 使用 IK 后端。
- 保留 Odin ROS 驱动及 Apache-2.0 许可证、五份比赛路线、抽样 PCD 和 Orin 触控屏 UI。
- 排除嵌套 Git、缓存、日志、备份、候选模型、开发草稿和重复地图工具。
- 原始备份缺少配置所引用的 Odin `1hao.bin`,因此重定位模式仍需从比赛设备补回该外部资产;纯里程计模式不受此限制。
- 自研 ROS 包仍保留原工程的 `Proprietary` 清单字段,公开到 GitHub 前必须由权利人统一选择开源许可证。
+6 -4
View File
@@ -9,7 +9,9 @@
└─ real/
├─ ik_real/ # IK 轨迹与早期真机控制
├─ sim2real/ # 第一代 Python 策略真机部署
─ sim2real_ros2/ # 最终比赛 ROS 2/C++ 真机部署
─ sim2real_v2/ # Python Sim2Real v2
├─ sim2real_ros2/ # ROS 2/C++ Sim2Real 初版
└─ sim2real_ros2_v2/ # ROS 2 导航原型及后续演进
```
## 数据流
@@ -24,14 +26,14 @@ MJCF + mjlab task
|
+----> Sim2Sim 策略验证
|
+----> Python Sim2Real ----> 电机 / IMU
+----> Python Sim2Real / v2 ----> 电机 / IMU
|
+----> ROS 2/C++ Sim2Real -> CAN / Odin / 导航 / 屏幕
+----> ROS 2/C++ Sim2Real -----> CAN / IMU / 导航
IK real --------------------------------> 电机
```
`rc_mjlab` 是自包含工程。训练、MJCF、MuJoCo、Sim2Sim、导航工具和策略权重通过相对路径绑定,因此保留其内部布局,没有为了目录外观拆散。第一代完整闭环见 `v0.3.0`,第一份新版 MJCF 与训练框架见 `v0.4.0`,随机化增强版见 `v0.5.0`,比赛最终训练架构见 `v0.6.0`,后期 MuJoCo 工具集见 `v0.7.0`,后期 Sim2Sim 与比赛 Rough 策略见 `v0.8.0`,完整导航打点工具见 `v0.8.1`最终比赛 ROS 2 部署见 `v0.9.0`
`rc_mjlab` 是自包含工程。训练、MJCF、MuJoCo、Sim2Sim、导航工具和策略权重通过相对路径绑定,因此保留其内部布局,没有为了目录外观拆散。第一代完整闭环见 `v0.3.0`,第一份新版 MJCF 与训练框架见 `v0.4.0`,随机化增强版见 `v0.5.0`,比赛最终训练架构见 `v0.6.0`,后期 MuJoCo 工具集见 `v0.7.0`,后期 Sim2Sim 与比赛 Rough 策略见 `v0.8.0`,完整导航打点工具见 `v0.8.1`Python Sim2Real v2 对应 `v0.9.0`ROS 2/C++ 初版对应 `v0.10.0`,简单导航原型对应 `v0.11.0`,完整 Odin/TensorRT 与站姿调参对应 `v0.11.1`,纯里程计导航联调对应 `v0.12.0`
详细说明见:
+17 -9
View File
@@ -1,6 +1,6 @@
# 真机控制与部署
# 真机控制版本演进
本目录保存 16DOF 轮足机器人从早期接口验证到最终比赛 ROS 2 部署的演进。
本目录保存 16DOF 轮足机器人从早期 Python 闭环到 ROS 2 部署的真机控制演进。
## `ik_real`
@@ -20,17 +20,25 @@
部署说明见 [`sim2real/README.md`](sim2real/README.md) 与 [`sim2real/DEPLOYMENT.md`](sim2real/DEPLOYMENT.md)。
## `sim2real_v2`
Python Sim2Real v2,保留 `53D -> 16D` 策略接口,并增加电机反馈新鲜度、Odin odom 诊断、命令平滑、Web 运行时诊断和安全监控工具。该版本对应重排主线的 `v0.9.0`
部署说明见 [`sim2real_v2/README.md`](sim2real_v2/README.md) 与 [`sim2real_v2/DEPLOYMENT.md`](sim2real_v2/DEPLOYMENT.md)。
## `sim2real_ros2`
`last_not_slalom_1050` 最终比赛工程的规范化归档,包含:
ROS 2/C++ Sim2Real 初版,将策略热路径迁移为 50 Hz C++ 推理和 200 Hz CAN 电机循环,并加入 ROS 2 消息、命令仲裁、Nav2 与统一启动结构。该版本对应重排主线的 `v0.10.0`
- ROS 2 Humble + C++ 运行时
- 50 Hz 策略推理与 200 Hz CAN 电机热路径
- Rough `model_6800`、Wall `model_84` 和 Crawl IK 模式
- Odin IMU/里程计驱动、简单导航、命令仲裁和触控屏 UI
- 比赛路线、抽样 PCD、Docker 与部署说明
原始快照没有随工程保存 Odin ROS 2 驱动源码,该依赖边界见 [`sim2real_ros2/README.md`](sim2real_ros2/README.md)。
`1050` 是比赛得分,不是模型编号。完整入口与缺失的 Odin 重定位地图边界见 [`sim2real_ros2/README.md`](sim2real_ros2/README.md)。
## `sim2real_ros2_v2`
ROS 2 Sim2Real v2 导航原型,在初版基础上增加简单导航节点、PCD 交互定位、任务点/任务序列和 Web 导航调试。该版本对应重排主线的 `v0.11.0`
`v0.11.1` 在同一目录继续演进,首次随工程归档完整 Odin 驱动、TensorRT、多策略切换和硬件诊断,并使用 `hip=0.670``knee=-1.390` 的调参站姿。各 Tag 可恢复对应阶段,当前目录说明见 [`sim2real_ros2_v2/README.md`](sim2real_ros2_v2/README.md)。
`v0.12.0` 继续在同一目录保存 odom 快照,固定纯里程计模式,加入 odom fallback 的 TF 冲突保护、A_min 路线和多地图工具;默认 Rough 策略为 `model_9600`,默认站姿回到比赛站姿。
## 实机记录
@@ -1,14 +1,6 @@
build/
install/
log/
logs_v2_web/
map/load/
src/odin_ros_driver/log/
src/odin_ros_driver/recorddata/
src/odin_ros_driver/image/
*.bak_*
__pycache__/
*.py[cod]
.colcon/
.vscode/
compile_commands.json
@@ -156,7 +156,7 @@ sudo udevadm trigger
统一启动文件 `sim2real_system.launch.py` 支持模块化激活传感器驱动和 Nav2 导航栈:
* `launch_driver`(默认:`true`):启动 `odin_ros_driver` 节点以获取 IMU 和点云遥测。
* `launch_nav2`(默认:`false`):按需启动 ROS2 Navigation2;比赛默认使用 `simple_nav_node.py` 的路线跟踪
* `launch_nav2`(默认:`true`):启动 ROS2 Navigation2 规划器、控制器、costmap、AMCL 和 pointcloud_to_laserscan
#### 1. 完整真实硬件闭环(默认)
启动运动控制运行时、物理 CAN 桥接、Odin 传感器驱动和 Nav2 导航:
@@ -175,3 +175,4 @@ ros2 launch sim2real_bringup sim2real_system.launch.py dry_run:=true launch_driv
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_driver:=false launch_nav2:=false
```
@@ -62,7 +62,6 @@ COPY src/sim2real_nav2 sim2real_nav2
# 拷贝策略文件与运行脚本
WORKDIR /sim2real_ws
COPY policies policies
COPY map map
COPY start_sim2real.sh start_sim2real.sh
RUN chmod +x start_sim2real.sh
+60 -90
View File
@@ -1,104 +1,74 @@
# ROS 2 最终比赛 Sim2Real
# ROS 2/C++ Sim2Real 初版
本目录归档 `last_not_slalom_1050` 真机工程,对应 RC_WheelLeg 在 RoboCon 仿生足式障碍赛使用的最终 ROS 2 部署栈。`1050` 是比赛得分,不是模型编号;比赛 Rough 策略为 `model_6800.onnx`
本目录归档 `real/sim2real_ros2`,对应重排主线的 `v0.10.0`。这是轮腿机器人 Sim2Real 部署栈从 Python 运行时迁移到 ROS 2 + C++ 的第一版系统工程
该里程碑计划标记为 `v0.9.0`。训练架构和策略来源见 `v0.6.0`,比赛 Rough 模型首次归档见 `v0.8.0`,导航打点与路线演进见 `v0.8.1`
本工程保留当前 `sim2real` 已验证的部署契约,同时将运行时热路径迁移到 C++:
## 系统闭环
- `53D` 策略观测契约不变
- `16D` 动作契约不变
- `50Hz` 策略循环与训练对齐
- `200Hz` 电机循环为专用 C++ 热路径
- ROS 2 作为导航、TF、诊断和启动管理的系统集成层
## 工作区布局
- `src/sim2real_interfaces`
硬件桥接与策略运行时共享的 ROS 2 消息定义。
- `src/sim2real_common`
共享常量、部署契约辅助函数、Mahony 姿态滤波器、站立平衡控制器、安全监控。
- `src/sim2real_hw`
面向硬件的桥接节点:RobStride CAN 收发、IMU/Odin 数据采集、看门狗、状态发布。
- `src/sim2real_runtime`
策略运行时节点:`53D→16D` ONNX 推理、命令滤波/仲裁、目标发布。
同时包含 `odom_relay_node`(里程计中继与 TF 广播)。
- `src/sim2real_nav2`
ROS 2 Navigation2 (Nav2) 配置包:参数、启动文件、AMCL、costmap、planner/controller。
- `src/sim2real_bringup`
统一启动文件与运行时参数配置。
- `src/odin_ros_driver`
Odin 传感器 ROS 2 驱动(含 IMU、点云、里程计发布)。
- `docs`
架构说明与迁移计划。
## 目标架构
```text
Odin IMU / Odom ──> hardware bridge ──> RuntimeState
|
导航 / 遥控 / 屏幕 ──> cmd mux ──> policy runtime (50 Hz)
|
RuntimeTarget
|
hardware bridge / CAN (200 Hz)
Odin / IMU / Odom ---> sim2real_hw ---> sim2real_runtime ---> sim2real_hw
| | |
v v v
RuntimeState RuntimeTarget 电机 CAN 指令
| |
+-------> 诊断 / 遥测
Nav2 / cmd_vel ------------------------------> sim2real_runtime
(经 odom_relay_node 提供 odom→base_link TF)
```
核心约束:
## 当前状态
- 53 维策略观测、16 维动作输出。
- Rough`model_6800`,优先 TensorRT,失败时回退 ONNX Runtime。
- Wall`model_84`,同样保留 TensorRT 与 ONNX 两种文件。
- Crawl:比赛配置使用解析 IK,不加载 Crawl RL 权重。
- 默认站姿:髋俯仰 `0.550`、膝关节 `-1.125`
- 默认命令源:`NAV`;默认定位模式:`relocal`
已完成 Phase 0-5 的全部迁移:
## 目录
1. ✅ 冻结部署契约(deployment_contract.hpp
2. ✅ ROS 2 包结构搭建
3. ✅ 硬件热路径迁移至 C++SocketCAN 驱动、200Hz 电机循环)
4. ✅ ONNX 策略运行时迁移至 C++(50Hz 推理循环)
5. ✅ 导航与诊断通过 ROS 2 接入(Nav2 + odom_relay + TF
```text
sim2real_ros2/
├─ src/
│ ├─ sim2real_interfaces/ # RuntimeState / RuntimeTarget 消息
│ ├─ sim2real_common/ # 部署契约、滤波、平衡和安全监控
│ ├─ sim2real_hw/ # SocketCAN、IMU 和 200 Hz 电机热路径
│ ├─ sim2real_runtime/ # 策略、命令仲裁、导航、Web API
│ ├─ sim2real_nav2/ # Nav2 配置入口
│ ├─ sim2real_bringup/ # 统一参数和启动文件
│ └─ odin_ros_driver/ # Odin ROS 驱动(Apache-2.0
├─ policies/ # 比赛实际使用的 Rough / Wall 模型
├─ map/ # 比赛路线和抽样 PCD
├─ screen/ # Orin 800×600 触控面板
├─ docs/ # 架构、遥控、Web 和迁移说明
├─ Dockerfile
└─ start_sim2real.sh
```
## 契约来源
## 构建与运行
迁移过程中以下文件被视为真值源:
目标环境是 Ubuntu 22.04、ROS 2 Humble 和 Jetson Orin。系统依赖和 Docker 流程见 [`DEPLOYMENT_GUIDE.md`](DEPLOYMENT_GUIDE.md)。
- `sim2real/deployment_manifest.yaml`
- `sim2real/interface/motor_mapping.py`
- `sim2real/interface/real_io.py`
- `sim2real/policy/policy_runner.py`
- `sim2real/web/session.py`
```bash
cd 05_software/real/sim2real_ros2
colcon build --merge-install --cmake-args -DCMAKE_BUILD_TYPE=Release
./start_sim2real.sh
```
## 注意事项
运行参数和模型/路线均使用工作区根目录相对路径,因此应从本目录启动。常用启动覆盖:
```bash
# 纯里程计模式,不等待 Odin 重定位地图
./start_sim2real.sh localization_mode:=odom \
odin_config_file:=src/odin_ros_driver/config/control_command_odom.yaml
# 禁止驱动,仅做软件链路检查
./start_sim2real.sh launch_driver:=false launch_remote:=false
```
## 必须补充的部署资产
最终源目录配置引用了 Odin `map/1hao.bin`,但工作区备份中不存在这个文件;全盘检索也未找到同名文件。为避免用来源不明的 `.bin` 冒充比赛地图,本仓库不伪造该资产。
使用 `relocal` 前必须:
1. 从比赛 Orin 或 Odin 建图备份取得真实 `1hao.bin`
2. 修改 `src/odin_ros_driver/config/control_command_relocal.yaml` 中的 `relocalization_map_abs_path` 为目标机绝对路径。
3. 核对文件哈希并在发布说明中补充来源。
缺少该文件时请使用 `localization_mode:=odom`,不要宣称重定位闭环已复现。地图和路线边界见 [`map/README.md`](map/README.md)。
## 归档边界
已保留:
- 最终六个 ROS 2 包、Odin 驱动源码、比赛设备标定参数和预编译 SDK 静态库。
- 最终 Rough/Wall ONNX 与比赛机 TensorRT engine。
- 五份最终工程路线、1 号场地抽样 PCD、屏幕 UI 和启动脚本。
- Odin 驱动 Apache-2.0 许可证。
未保留:
- 嵌套 `.git``__pycache__`、日志、备份、构建/安装目录。
- 未被比赛配置引用的候选模型与候选 TensorRT engine。
- 开发计划、任务草稿、重复地图工具和运行时轨迹。
- 原备份中大小为 0 的浏览器静态页面;HTTP JSON API 和屏幕 UI 源码仍保留。
TensorRT engine 与 JetPack、TensorRT 版本及 GPU 架构有关;其他机器应从同名 ONNX 重新生成,不应默认复用比赛 engine。模型哈希见 [`policies/README.md`](policies/README.md)。
## 安全与开源状态
- 真机运行前必须架空轮组验证 CAN 映射、方向、零位、急停和限幅。
- `deployment_contract.hpp` 是电机映射和动作缩放真值源;参考 YAML 不会自动修改 C++ 契约。
- 自研 ROS 包的 `package.xml` 仍保留原工程的 `Proprietary` 字段。迁移到 GitHub 公共开源前,需要由项目负责人选择许可证并统一修改;本次整理不代替权利人作许可证决定。
- 当前 Windows 环境只能做静态检查,不能证明 ROS 2、SocketCAN、Odin SDK 或 TensorRT 真机运行成功。
- 开发目标为 Linux + ROS 2 Humble,运行于 Jetson Orin / x86_64。
- Windows 仅作为编辑环境使用。
- 观测顺序、动作缩放、默认站姿、电机映射不得独立修改,
除非训练与部署同步更新。
- 原始快照中的 `src/odin_ros_driver` 是空目录,本版本仍需要另行提供兼容的 Odin ROS 2 驱动;其源码从后续版本开始随工程归档。
- 自研 ROS 包保留原始 `Proprietary` 清单字段,公开发布前仍需统一许可证和维护者信息。
@@ -21,7 +21,7 @@ src/sim2real_runtime/src/remote_uart_node.py
## 2. 通道映射
通道映射与本仓库第一代 Python Sim2Real 实现中的遥控器配置保持一致。
通道映射与前一阶段 Python Sim2Real 中的遥控器实现保持一致。
| 遥控器通道 | ROS 2 输出 | 含义 | 默认最大值 |
|---|---|---|---:|
@@ -64,10 +64,6 @@ remote_invert_vy: false
remote_invert_yaw: true
remote_publish_inactive_zero: true
remote_estop_latch: true
remote_estop_channel: 7
remote_estop_level: "high"
remote_estop_debounce_frames: 3
remote_estop_require_remote_mode: true
remote_poll_hz: 50.0
```
@@ -280,7 +276,7 @@ ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false
/safety/estop: true
```
由于当前 `remote_estop_latch: true`,急停是锁存式行为:`REMOTE` 模式下,CH7 连续 3 帧有效高位后,节点会发布急停,并保持内部急停已触发状态。恢复运行通常需要重启系统或手动发布复位信号,并确认机器人安全。
由于当前 `remote_estop_latch: true`,急停是锁存式行为:一旦 CH7 高位触发,节点会发布急停,并保持内部急停已触发状态。恢复运行通常需要重启系统或手动发布复位信号,并确认机器人安全。
### 8.4 机器人行为效果
@@ -1,164 +0,0 @@
# Pure Pursuit Yaw检查优化 - 最终配置
## 🎯 核心问题
**现象**: 机器人没转到位就开始前进 → 斜着撞杆子
**根本原因**: Pure Pursuit的yaw检查不够严格
---
## ✅ 已修改的参数
### runtime.yaml
```yaml
# 关键修改1: Script模式yaw门限
nav_slalom_script_yaw_gate_deg: 8.0 # 从12度 → 8度
# 关键修改2: 全局yaw容差
nav_goal_yaw_tolerance_deg: 8.0 # 从12度 → 8度
```
**作用**:
- yaw偏差 > 8度时:只转向,不前进
- yaw偏差 ≤ 8度时:才允许前进
---
## 📊 完整配置总结
### 1. 禁用干扰的规划器
```yaml
nav_local_planner_enabled: false
nav_astar_enabled: false
```
### 2. 速度和精度
```yaml
nav_slalom_max_vx: 0.50
nav_slalom_lookahead: 0.15
nav_slalom_tolerance: 0.05
```
### 3. Yaw控制(新增)
```yaml
nav_goal_yaw_tolerance_deg: 8.0
nav_slalom_script_yaw_gate_deg: 8.0
```
---
## 🔍 工作原理
### Pure Pursuit算法流程
```
1. 看前方lookahead距离的目标点
2. 计算到目标的方向和距离
3. 检查yaw偏差
- 如果 yaw偏差 > yaw_gate (8度):
只发wz转向,vx=0
- 如果 yaw偏差 ≤ yaw_gate (8度):
发vx前进 + wz微调
4. 到达目标点,推进下一个
```
### 之前的问题
```
yaw_gate = 12度(太宽松)
yaw偏差11度时就开始前进
还没对准就冲出去
斜着撞杆子
```
### 修改后
```
yaw_gate = 8度(更严格)
yaw偏差必须≤8度才前进
基本对准后才移动
不会斜着撞
```
---
## 📁 使用的文件
**路线**: `points_nav1007_optimized.json`
- 总航点: 47
- 绕杆航点: 21
- 参数: speed=0.50, lookahead=0.15, tolerance=0.05
**地形**: `tools/nav_tools/xml/A.xml`
**配置**: `runtime.yaml` (已修改)
---
## 🚀 下一步
### 实机测试
```bash
# 1. 重启系统加载新配置
ros2 launch sim2real_bringup sim2real_system.launch.py
# 2. 验证参数
ros2 param get /sim2real_simple_nav_node nav_slalom_script_yaw_gate_deg
# 应该显示: 8.0
ros2 param get /sim2real_simple_nav_node nav_goal_yaw_tolerance_deg
# 应该显示: 8.0
# 3. 加载路线
# 使用: sim2real_ros2_v2_ooo/map/routes/points_nav1007_optimized.json
# 4. 监控
ros2 topic echo /cmd_vel_nav
# 观察: 转向时vx应该接近0,对准后才有vx速度
```
---
## 💡 预期效果
**修改前**:
- 机器人边转边进
- yaw偏差大时仍有前进速度
- 导致斜着撞杆子
**修改后**:
- 转向时几乎不前进(vx≈0
- 对准后才快速前进(vx=0.5
- 动作分离:先转向,后前进
---
## ⚠️ 如果还有问题
### 场景A: 还是斜着撞
可能需要进一步收紧:
```yaml
nav_slalom_script_yaw_gate_deg: 5.0 # 改为5度
```
### 场景B: 太慢,一直在转
说明yaw_gate太严格:
```yaml
nav_slalom_script_yaw_gate_deg: 10.0 # 放宽到10度
```
### 场景C: 卡顿
检查是否在等待yaw对准:
```bash
ros2 topic echo /simple_nav/status
# 看是否一直在"aligning"状态
```
---
**配置已优化完成,ready for testing!** 🎯
@@ -0,0 +1,20 @@
Tcl_0: [-0.009160, -0.999960, 0.000320, 0.032150,
0.002390, -0.000340, -1.000000, -0.011850,
0.999960, -0.009160, 0.002390, 0.005360,
0.000000, 0.000000, 0.000000, 1.000000]
cam_0:
image_width: 1600
image_height: 1296
k2: 0.000656
k3: -0.028961
k4: 0.045390
k5: -0.064513
k6: 0.038735
k7: -0.009903
p1: 0.000000
p2: 0.000000
A11: 736.894262
A12: -0.161150
A22: 736.611354
u0: 806.125535
v0: 639.650710
@@ -1,11 +0,0 @@
# 比赛地图与路线
`routes/` 保存 `last_not_slalom_1050` 工程中的五份路线快照,保留原文件名以维持运行时配置和版本演进关系。默认运行路线是 `routes/1hao_reall.json`
`1hao.pcd``v0.8.1` 导航工具中同一份抽样点云,包含 199,215 点、大小 9,876,010 字节,SHA-256 为 `48B231C52BECA51316F352300C8B2046133E92359E0855227D93DEB0D927AD34`。它用于本地规划和路线显示,不替代原始高密度点云。
## 缺失的 Odin 重定位地图
比赛配置需要 Odin 专用二进制地图 `1hao.bin`,但源备份没有该文件。源目录中另有两个名称和时间不同的 `.bin`,无法证明它们就是比赛使用地图,因此没有复制或重命名。
重定位部署时应从比赛设备取回真实文件,并把 `control_command_relocal.yaml` 中的占位绝对路径改为目标机实际位置。纯里程计模式不需要该文件。
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
@@ -1,14 +0,0 @@
# 比赛部署策略
这里只保留最终运行配置实际使用的 Rough 和 Wall 策略。Crawl 在比赛配置中使用 IK 后端,因此不归档未使用的 Crawl RL engine。
| 用途 | 文件 | SHA-256 |
| --- | --- | --- |
| Rough ONNX | `model_6800.onnx` | `3C994BDD3434AD15770A52AC0E8D229F502F00D6511CDD42C2E2C742301AEF13` |
| Rough TensorRT | `model_6800_fp16.engine` | `BDC6583AFD594E84D9A4AAA4BB2EEE279C7EBF207A57CF4802A2DF5309C5F663` |
| Wall ONNX | `model_84.onnx` | `E3A447782DF6C6E11E66C3ACBE41697FF88EBF5E75DCAB65EFCF9A2EB1CA639F` |
| Wall TensorRT | `model_84_fp16.engine` | `0FE2001113B7146B589D7A62A6E54BAD68574765EBA45782FB41A26F65A8721D` |
`model_6800.onnx``05_software/train/rc_mjlab/model_6800.onnx` 哈希一致。此处重复保留是为了让真机工作区可以独立部署。
TensorRT 文件只作为比赛机历史产物;更换 JetPack、TensorRT 或 GPU 后应从 ONNX 重新构建并重新核对数值误差。
@@ -1,22 +0,0 @@
# Orin 触控屏控制面板
`fullscreen_quit.py` 是比赛 Orin 外接 `800×600` 屏幕使用的控制面板,通过本机 `http://127.0.0.1:18080/api/*` 调用 ROS 2 Web bridge,不建立第二套控制协议。
```bash
cd <sim2real_ros2工作区>
DISPLAY=:0 python3 screen/fullscreen_quit.py
```
程序默认从脚本父目录自动确定工作区,也可设置 `SIM2REAL_REPO_DIR` 覆盖。显示权限检查:
```bash
python3 screen/check_display.py
```
自启动脚本:
- `install_autostart.sh`:图形桌面登录后启动。
- `install_boot_service.sh <用户名>`:安装 systemd 服务。
- `uninstall_boot_service.sh`:卸载服务。
使用屏幕启动系统前,先完成工作区构建、Odin 配置、CAN 映射检查和急停验证。
@@ -1,114 +0,0 @@
#!/usr/bin/env python3
"""
Print display-related diagnostics for launching the fullscreen quit screen.
"""
import glob
import getpass
import os
import shlex
import subprocess
from typing import List, Optional
def run(command: List[str]) -> str:
try:
result = subprocess.run(
command,
check=False,
capture_output=True,
text=True,
)
except OSError as exc:
return f"<failed: {exc}>"
output = result.stdout.strip()
error = result.stderr.strip()
if error:
return f"{output}\n{error}".strip()
return output
def current_user() -> str:
try:
return getpass.getuser()
except OSError:
return os.environ.get("USER", "unknown")
def xauth_from_x_processes() -> List[str]:
output = run(["ps", "-eo", "args"])
candidates = []
for line in output.splitlines():
if "Xorg" not in line and "Xwayland" not in line:
continue
try:
parts = shlex.split(line)
except ValueError:
parts = line.split()
for index, part in enumerate(parts[:-1]):
if part == "-auth":
candidates.append(parts[index + 1])
return candidates
def print_path_status(label: str, path: Optional[str]) -> None:
if not path:
print(f"{label}: <unset>")
return
exists = os.path.exists(path)
readable = os.access(path, os.R_OK) if exists else False
print(f"{label}: {path} exists={exists} readable={readable}")
def main() -> int:
uid = os.getuid()
print(f"user: {run(['whoami'])}")
print(f"uid: {uid}")
print(f"DISPLAY: {os.environ.get('DISPLAY', '<unset>')}")
print(f"XAUTHORITY: {os.environ.get('XAUTHORITY', '<unset>')}")
print()
print("candidate XAUTHORITY files:")
candidates = [
os.environ.get("XAUTHORITY"),
*xauth_from_x_processes(),
os.path.expanduser("~/.Xauthority"),
f"/run/user/{uid}/gdm/Xauthority",
f"/run/user/{uid}/Xauthority",
*glob.glob(f"/run/user/{uid}/*Xauthority*"),
]
seen = set()
for candidate in candidates:
key = candidate or "<unset>"
if key in seen:
continue
seen.add(key)
print_path_status(" -", candidate)
print()
print("X server processes:")
x_lines = [
line
for line in run(["ps", "-eo", "user,args"]).splitlines()
if "Xorg" in line or "Xwayland" in line
]
if x_lines:
for line in x_lines:
print(f" {line}")
else:
print(" <none found>")
print()
print("recommended SSH launch:")
print(" cd <workspace>/screen")
print(" bash run_fullscreen_quit.sh")
print()
print("if authorization fails, run this on the Nano desktop terminal once:")
print(f" DISPLAY=:0 xhost +SI:localuser:{current_user()}")
return 0
if __name__ == "__main__":
raise SystemExit(main())
File diff suppressed because it is too large Load Diff
@@ -1,27 +0,0 @@
#!/usr/bin/env bash
set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
AUTOSTART_DIR="$HOME/.config/autostart"
DESKTOP_FILE="$AUTOSTART_DIR/sim2real-screen.desktop"
mkdir -p "$AUTOSTART_DIR"
chmod +x "$SCRIPT_DIR/run_fullscreen_quit.sh"
cat > "$DESKTOP_FILE" <<EOF
[Desktop Entry]
Type=Application
Name=sim2real Screen
Comment=Start the sim2real touch control panel
Exec=$SCRIPT_DIR/run_fullscreen_quit.sh
Path=$SCRIPT_DIR
Terminal=false
X-GNOME-Autostart-enabled=true
StartupNotify=false
EOF
echo "Installed desktop autostart:"
echo " $DESKTOP_FILE"
echo
echo "It will start after this user logs into the graphical desktop."
echo "For boot-time use, enable automatic login for this user on the Orin desktop."
@@ -1,73 +0,0 @@
#!/usr/bin/env bash
set -euo pipefail
if ! command -v systemctl >/dev/null 2>&1; then
echo "systemctl not found; this installer is for systemd-based Linux." >&2
exit 1
fi
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
SERVICE_NAME="sim2real-screen.service"
SERVICE_PATH="/etc/systemd/system/$SERVICE_NAME"
if [ "${EUID:-$(id -u)}" -eq 0 ]; then
RUN_USER="${1:-${SUDO_USER:-rc2}}"
else
RUN_USER="${1:-$(id -un)}"
fi
if ! id "$RUN_USER" >/dev/null 2>&1; then
echo "User '$RUN_USER' does not exist. Usage: bash install_boot_service.sh rc2" >&2
exit 1
fi
RUN_GROUP="$(id -gn "$RUN_USER")"
RUN_HOME="$(getent passwd "$RUN_USER" | cut -d: -f6)"
AUTOSTART_FILE="$RUN_HOME/.config/autostart/sim2real-screen.desktop"
chmod +x "$SCRIPT_DIR/run_fullscreen_quit.sh" "$SCRIPT_DIR/run_boot_screen_service.sh"
if [ -f "$AUTOSTART_FILE" ]; then
rm -f "$AUTOSTART_FILE"
echo "Removed desktop autostart to avoid duplicate screen instances:"
echo " $AUTOSTART_FILE"
fi
SERVICE_CONTENT="[Unit]
Description=sim2real touchscreen control panel
Wants=display-manager.service
After=systemd-user-sessions.service display-manager.service
StartLimitIntervalSec=0
[Service]
Type=simple
User=$RUN_USER
Group=$RUN_GROUP
WorkingDirectory=$SCRIPT_DIR
Environment=HOME=$RUN_HOME
Environment=DISPLAY=:0
Environment=PYTHONUNBUFFERED=1
ExecStart=$SCRIPT_DIR/run_boot_screen_service.sh
Restart=always
RestartSec=2
KillSignal=SIGINT
TimeoutStopSec=20
[Install]
WantedBy=graphical.target
"
printf "%s" "$SERVICE_CONTENT" | sudo tee "$SERVICE_PATH" >/dev/null
sudo systemctl daemon-reload
sudo systemctl enable "$SERVICE_NAME"
echo "Installed and enabled:"
echo " $SERVICE_PATH"
echo
echo "Start now:"
echo " sudo systemctl restart $SERVICE_NAME"
echo
echo "Check status/logs:"
echo " systemctl status $SERVICE_NAME --no-pager"
echo " journalctl -u $SERVICE_NAME -f"
echo
echo "Important: the graphical desktop for user '$RUN_USER' must auto-login, otherwise Tk cannot open DISPLAY=:0."
@@ -1,23 +0,0 @@
#!/usr/bin/env bash
set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
export DISPLAY="${DISPLAY:-:0}"
export PYTHONUNBUFFERED=1
echo "[sim2real-screen] service starting as $(id -un), DISPLAY=$DISPLAY"
for _ in $(seq 1 120); do
if [ -S "/tmp/.X11-unix/X${DISPLAY#:}" ]; then
break
fi
sleep 1
done
if [ ! -S "/tmp/.X11-unix/X${DISPLAY#:}" ]; then
echo "[sim2real-screen] X11 socket for DISPLAY=$DISPLAY not found" >&2
exit 1
fi
exec "$SCRIPT_DIR/run_fullscreen_quit.sh"
@@ -1,24 +0,0 @@
#!/usr/bin/env bash
set -u
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
cd "$SCRIPT_DIR"
export DISPLAY="${DISPLAY:-:0}"
if [ -z "${XAUTHORITY:-}" ]; then
UID_VALUE="$(id -u)"
for candidate in \
"$HOME/.Xauthority" \
"/run/user/$UID_VALUE/gdm/Xauthority" \
"/run/user/$UID_VALUE/Xauthority"
do
if [ -f "$candidate" ]; then
export XAUTHORITY="$candidate"
break
fi
done
fi
python3 fullscreen_quit.py
Binary file not shown.

Before

Width:  |  Height:  |  Size: 34 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 35 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 27 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 27 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 35 KiB

@@ -1,11 +0,0 @@
#!/usr/bin/env bash
set -euo pipefail
SERVICE_NAME="sim2real-screen.service"
SERVICE_PATH="/etc/systemd/system/$SERVICE_NAME"
sudo systemctl disable --now "$SERVICE_NAME" 2>/dev/null || true
sudo rm -f "$SERVICE_PATH"
sudo systemctl daemon-reload
echo "Removed $SERVICE_NAME"
@@ -1,13 +0,0 @@
v0.11.0 2026_0618
Required Minimum Firmware Version0.12.0
1.Reduce cloud slam latency
2.Improve get mapping result
v0.10.5 2026_0525
1.modify recorddata formatadd device_id、algorithm_version key
2.fix download map fail when in slam mode(USB2.0)
3.fix upload map fail when in relocalization mode(USB2.0)
v0.10.4 2026_0522
1.fix usb2.0 heartBeat timeout to cause soft detaching
2.add custom_init_pose_search_radius and custom_init_pose_max_rot_deg in control_command.yaml
@@ -1,821 +1,5 @@
# Odin_ROS_Driver Readme
# Odin 驱动依赖占位
ROS driver suite for Odin sensor modules (Manifold Tech Ltd.)
`real/sim2real_ros2` 原始快照中的 `src/odin_ros_driver` 为空目录,但启动文件、Dockerfile 和 `sim2real_bringup` 已经引用该包。
Odin1 wiki: https://manifoldtechltd.github.io/wiki/Odin1/Cover.html
## Odin_ROS_Driver
Compatibility:
● ROS 1(LTS Release: Noetic recommended)
● ROS 2(LTS Release: Humble recommended)
## Important Notice:
This driver package provides core functionality for point cloud SLAM applications and targets specific use cases. It is intended exclusively for technical professionals conducting secondary development. End users must perform scenario-specific optimization and custom development to align with operational requirements in practical deployment environments.
## 1. Version
Current version: v0.12.0
Required device firmware version: v0.12.0
## 2. Preparation
### 2.1 OS Requirement
● Ubuntu 20.04 for ROS Noetic and ROS2 Foxy;
● Ubuntu 22.04 for ROS2 Humble;
● Ubuntu 18.04 is currently not supported;
● Ubuntu 24.04 is not officially supported but may work with some modifications.
### 2.2 Dependencies
● Opencv >= 4.2.0(recommand 4.5.5/4.8.0. Make sure only one version of opencv is installed)
● yaml-cpp
● thread
● OpenSSL
● Eigen3
### 2.3 Dependencies Install
#### 2.3.1 System
```shell
sudo apt update
sudo apt-get install build-essential cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev
```
#### 2.3.2 yaml-cpp
```shell
sudo apt update
sudo apt install -y libyaml-cpp-dev
```
#### 2.3.3 libusb
```shell
sudo apt update
sudo apt install -y libusb-1.0-0-dev
```
#### 2.3.4 opencv
```shell
sudo apt update
sudo apt-get install libopencv-dev
```
#### 2.3.4 ROS install
For ROS Noetic installation, please refer to:
[ROS Noetic installation instructions](https://wiki.ros.org/noetic/Installation)
For ROS2 Foxy installation, please refer to:
[ROS Foxy installation instructions](https://docs.ros.org/en/foxy/Installation/Ubuntu-Install-Debians.html)
For ROS2 Humble installation, please refer to:
[ROS Humble installation instructions](https://docs.ros.org/en/humble/Installation/Ubuntu-Install-Debians.html)
## 3. Preparation
### 3.1 Create Udev rules
```shell
sudo vim /etc/udev/rules.d/99-odin-usb.rules
```
Add the following content to the 99-odin-usb.rules file
```shell
SUBSYSTEM=="usb", ATTR{idVendor}=="2207", ATTR{idProduct}=="0019", MODE="0666", GROUP="plugdev"
```
Reload rules and reinsert devices
```shell
sudo udevadm control --reload
sudo udevadm trigger
```
### 3.2 OS Requirement
```shell
git clone https://github.com/manifoldsdk/odin_ros_driver.git catkin_ws/src/odin_ros_driver
```
Note:
Please clone the source code into the "[ros_workspace]/src/" folder, otherwise compilation errors will occur.
### 3.3 make
#### 3.3.1 ROS1 (Noetic for example):
```shell
source /opt/ros/noetic/setup.bash
./script/build_ros.sh
```
#### 3.3.2 ROS2 (Foxy for example):
```shell
source /opt/ros/foxy/setup.bash
./script/build_ros2.sh
```
### 3.4 run:
#### 3.4.1 ROS1 (Noetic for example):
```shell
source [ros_workspace]/devel/setup.bash
roslaunch odin_ros_driver [launch file]
```
● odin_ros_driver: package name;
● launch file: launch file;
● ros_workspace: User's ROS environment workspace;
```shell
roslaunch odin_ros_driver odin1_ros1.launch
```
#### 3.4.2 ROS2 (Foxy for example):
```shell
source [ros2_workspace]/install/setup.bash
ros2 launch odin_ros_driver [launch file]
```
● odin_ros_driver: package name;
● launch file: launch file;
● ros2_workspace: User's ROS2 environment workspace;
ROS2 Demo Launch Instructions:
```shell
ros2 launch odin_ros_driver odin1_ros2.launch.py
```
### 3.5 Operation Mode:
The operation mode can be configured via the `custom_map_mode` parameter in config/control_command.yaml.
#### Odometry mode
Set `custom_map_mode = 0` to enable odometry mode. In this mode, the map frame and odom frame share the same pose.
If the odom data is found to drift, the script command "./set_param.sh algo_reset 1" can be used to dynamically reset the algorithm.
#### SLAM mode
Set `custom_map_mode = 1` to enable slam mode. This mode provides a complete SLAM system that builds upon the Odometry Mode by adding **loop closure detection** and **map saving** capabilities.
After launching the driver, odin1 will automatically perform mapping and cache map data. When the scene capture is complete, users need to execute `./set_param.sh save_map 1` in the driver's source directory to save all map data collected since the program started. The map will be saved to the location specified by the `mapping_result_dest_dir` and `mapping_result_file_name` parameters in config/control_command.yaml. If these parameters are not specified, default values will be used.
After the initial save, you can execute the command again to save a new map. Each save operation will generate a new map file. (Please allow at least 5 seconds between consecutive save operations)
The map origin corresponds to the odom coordinate system's origin at the program's startup.
##### Relocalization mode
To enable relocalization, set `custom_map_mode = 2` and specify the absolute path to the pre-built map using the `relocalization_map_abs_path` parameter in config/control_command.yaml.
Once launched, odin1 will initiate the relocalization process based on the current viewpoint and the specified map. To ensure a high success rate, it is recommended to starting within 1 meter ±10 degrees of the original position and orientation from the SLAM trajectory.
Note that relocalization performance is highly environment-dependent. In highly distinctive scenes, successful matching may occur even beyond the 1m/10° range, while other environments may require more stringent conditions. We advise testing in your target environment to determine practical tolerances.
If relocalization fails initially, the system will temporarily operate in a fallback SLAM mode (map saving is disabled in this state). During this time, you can freely move odin1. It will continue relocalization attempts in the background. Once successful, the TF between map and odom frames will be published. (Tip: Gently shaking or moving the device after initialization can help improve relocalization accuracy.)
The following topics are published in the odom frame: `/odin1/cloud_slam, /odin1/odom, /odin1/highodom and /odin1/path`. To obtain these in the map frame, apply the TF from odom frame to map frame.
## 4. File structure and data format
### 4.1 File structure
```shell
Odin_ROS_Driver/ // ROS1/ROS2 driver package
3rdparty/ // Third-party libraries
src/
host_sdk_sample.cpp // Example source code
yaml_parser.cpp // Source code for reading yaml parameters
rawCloudRender.cpp // Source code for RenderCloud
depth_image_ros_node.cpp //depth_image_ros_node
depth_image_ros2_node.cpp //depth_image_ros2_node
pcd2depth_ros.cpp //Source code for pcd2depth_ros
pcd2depth_ros2.cpp //Source code for pcd2depth_ros2
pointcloud_depth_converter.cpp //Source code for pointcloud_depth_converter
cloud_reprojection_ros.cpp //Source code for cloud reprojection node (ROS1/ROS2)
cloud_reprojector.cpp //Core logic for cloud reprojection
lib/
liblydHostApi_amd.a // Static library for AMD platform
liblydHostApi_arm.a // Static library for ARM platform
include/
host_sdk_sample.h // Example header file
lidar_api_type.h // API data structure header file
lidar_api.h // API function declarations
yaml_parser.h // Parameter file reading header file
rawCloudRender.h // API about RenderCloud
data_logger.h // LOG about save_data
depth_image_ros_node.hpp // depth_image_ros_node
depth_image_ros2_node.hpp // depth_image_ros2_node
pointcloud_depth_converter.hpp // pointcloud_depth_convert
cloud_reprojection_ros_node.hpp // cloud_reprojection_ros_node (ROS1/ROS2)
cloud_reprojector.hpp // Core class for cloud reprojection
config/
control_command.yaml // Control parameter file for driver
calib.yaml // Machine calibration yamldiffer for each individual device. Retrieved from the device everytime it connects to ROS driver
launch_ROS1/
odin1_ros1.launch // ROS1 launch file
launch_ROS2/
odin1_ros2.launch.py // ROS2 launch file
script/
build_ros1.sh // Installation script for ROS1
build_ros2.sh // Installation script for ROS2
recorddata/ // holds recorded data that can import into MindCloud
log/ // holds log files
Driver_{timestamp}/ // holds all log folders for each time driver started
Conn_{timestamp}/ // holds all log files for each odin1 device connection
dev_status.csv // device status log file
README.md // Usage instructions
CMakeLists.txt // CMake build file
License // License file
```
### 4.2 File structure
| Launch File Name | Description |
|--------------------------|-------------|
| odin1_ros1.launch | Launch file for ROS1 - Odin1 Basic Operations Demo |
| odin1_ros2.launch.py | Launch file for ROS2 - Odin1 Basic Operations Demo |
### 4.3 ROS topics
Internal parameters of the Odin ROS driver are defined in config/control_command.yaml. Below are descriptions of the commonly used parameters:
| Topic |control_command.yaml | Detailed Description |
|---------------------|----------------------|----------------------|
| odin1/imu | sendimu | Imu Topic |
| odin1/image | sendrgb | RGB Camera Topic, decoded from original jpeg data from device, bgr8 format |
| odin1/image_undistort | sendrgbundistort | undistorted RGB Camera Topic, processed with calib.yaml from device |
| odin1/image/compressed | sendrgbcompressed | RGB Camera compressed Topic, original jpeg data from device |
| odin1/cloud_raw | senddtof | Raw_Cloud Topic |
| odin1/cloud_render | sendcloudrender | Render_Cloud Topic, processed with raw point cloud, rgb image, and calib.yaml from device |
| odin1/cloud_slam | sendcloudslam | Slam_PointCloud Topic |
| odin1/odometry | sendodom | Odom Topic |
| odin1/odometry_high | sendodom | high frequency Odom Topic |
| odin1/path | showpath | Odom Path Topic |
| tf | sendodom | tf tree Topic |
| odin1/depth_img_competetion | senddepth | Dense depth image Topic. Demo, high computing power required. One-to-one with odin1/image_undistort. To utilize the data please directly subscribe to this topic instead of echoing it. Original value is already depth data, no need for further convert. |
| odin1/depth_img_competetion_cloud | senddepth | Dense Depth_Cloud Topic. Demo, high computing power required |
| odin1/reprojected_image | sendreprojection | Reprojected cloud to image Topic. Projects cloud_slam to camera image using odometry. Processed on host device. |
### 4.4 Data format
1. The raw point cloud (cloud_raw) has the following fields:
```
float32 x // X axis, in meters
float32 y // Y axis, in meters
float32 z // Z axis, in meters
uint8 intensity // Reflectivity, range 0255
uint16 confidence // Point confidence, actual value range from 0 to around 1300 in typical scene, higher value means more reliable. Recommanded filtering threshold is 30-35, should be adjusted accordingly.
float32 offset_time // Time offset relative to the base timestamp unit: s
```
To work with this custom format in PCL, first define the point type:
```cpp
/*** LS ***/
namespace ls_ros {
struct EIGEN_ALIGN16 Point {
float x;
float y;
float z;
uint8_t intensity;
uint16_t confidence;
float offset_time;
EIGEN_MAKE_ALIGNED_OPERATOR_NEW
};
} // namespace ls_ros
POINT_CLOUD_REGISTER_POINT_STRUCT(ls_ros::Point,
(float, x, x)
(float, y, y)
(float, z, z)
(uint8_t, intensity, intensity)
(uint16_t, confidence, confidence)
(float offset_time , offset_time)
)
```
Then, you can easily convert a ROS sensor_msgs::PointCloud2 message into a PCL point cloud:
```
pcl::PointCloud<ls_ros::Point> ls_cloud;
pcl::fromROSMsg(*msg, ls_cloud);
```
2. The slam point cloud (cloud_slam) and directly rendered point cloud (cloud_render) has the following fields:
```
float32 x // X axis, in meters
float32 y // Y axis, in meters
float32 z // Z axis, in meters
float32 rgb // RGB value
```
### 4.5 Other functionalities
|control_command.yaml | Detailed Description |
|-----------------------|----------------------|
| use_host_ros_time | Time synchronization mode: 0 - use odin internal system time as data timestamp (typical and recommended); 1 - use host ROS time upon receive (not recommended for most users); 2 - align odin1 time to host time via NTP-like synchronization, timestamp is the sensor data reception time on host time axis. |
| strict_usb3.0_check | Strict USB3.0 check, if off, allow connection even if usb connection is below usb 3.0 |
| recorddata | Record data in specific format that can be imported into MindCloud(TM) for post-processing. Please be aware that this will consume a lot of storage space. Testing shows 9.5G for 10mins of data. The per-frame timestamps written into the recorded files (IMU / image / point cloud / pose / rotate) follow the same alignment policy as `use_host_ros_time`, so under NTP mode (`use_host_ros_time=1` or `2`) the recorded timestamps are NTP-aligned host time instead of odin1 boot time. <br>录制文件 (IMU / 图像 / 点云 / Pose / Rotate) 中每帧的时间戳与 `use_host_ros_time` 采用相同对齐策略:在 NTP 模式 (`use_host_ros_time=1``2`) 下,录制时间戳为 NTP 对齐后的主机时间,而非 odin1 开机时间。 |
| devstatuslog | Device status logging, currently save device status (soc temperature, cpu usage, ram usage, dtof sensor temp .etc) and data tx & rx rate to devstatus.csv under log folder. A new file will be created every time the driver is started. |
| showcamerapose | Display Camera Pose and Field of View. |
| custom_map_mode | Operation Modes: Mode 0 - Odometry mode: The map frame and odom frame share the same pose. Mode 1 - Mapping (with loop closure) mode: This mode supports map saving. Mode 2 - Relocalization mode: Requires specifying the absolute path to the map file. After successful relocalization, it will output the TF relationship between the map and odom frames.|
| custom_init_pos | Initialization Position (currently unused). |
| relocalization_map_abs_path | Absolute Path to Map File: Used for relocalization mode. |
| mapping_result_dest_dir and mapping_result_file_name| Path and Name for Saving Maps in Mapping Mode: If not specified, default values will be used. |
### 4.6 Runtime AE/AWB Tuning via ROS Service / 通过 ROS Service 在线调节 AE/AWB
The driver hosts four ROS services that let a side terminal tune the
camera's auto exposure (AE) and auto white balance (AWB) at runtime,
while the main data streams keep flowing. The same SDK call is shared
with the driver's main control path and serialised by an internal
mutex, so it is safe to invoke these services concurrently with normal
operation.
驱动启动后会注册 4 个 ROS Service,允许在不重启 driver 的前提下,从另一个终端动态调节
相机的自动曝光(AE)和自动白平衡(AWB)。底层 SDK 调用与驱动主控制路径共享同一把
互斥锁,因此可以与正常数据流并发调用。
**Service list / Service 一览**
| Service name | Type / 类型 | Purpose / 用途 |
|---|---|---|
| `/odin1/get_ae` | `odin_ros_driver/srv/GetAe` | Query current AE status / 查询当前 AE 状态 |
| `/odin1/get_awb` | `odin_ros_driver/srv/GetAwb` | Query current AWB status / 查询当前 AWB 状态 |
| `/odin1/set_ae` | `odin_ros_driver/srv/SetAe` | Set AE mode and (manual) exposure / gain / 设置 AE 模式和手动曝光/增益 |
| `/odin1/set_awb` | `odin_ros_driver/srv/SetAwb` | Set AWB mode and (manual) R/B gain / 设置 AWB 模式和手动 R/B 增益 |
#### 4.6.1 Request fields, ranges, physical meaning / 请求字段、范围与物理含义
**`SetAe.Request`**
| Field | Range / 范围 | Meaning / 含义 |
|---|---|---|
| `mode` | `0` (AUTO) or / 或 `1` (MANUAL) | `0` = device runs its own AE loop, the two floats below are ignored / 设备自动调 AE,下方参数被忽略<br>`1` = device locks AE and applies the provided values / 设备锁 AE 并应用提供的值 |
| `exposure_time` | `0.0001` ~ `0.033` s (manual only / 仅手动模式) | Sensor exposure time per frame. Longer = brighter but more motion blur / 每帧传感器曝光时间。越长越亮但运动模糊增大 |
| `gain` | `1.0` ~ `64.0` (manual only / 仅手动模式) | Analog gain. Higher = brighter output but worse SNR / 模拟增益。越大越亮但信噪比越差 |
**`SetAwb.Request`**
| Field | Range / 范围 | Meaning / 含义 |
|---|---|---|
| `mode` | `0` (AUTO) or / 或 `1` (MANUAL) | `0` = device runs its own AWB loop / 设备自动 AWB<br>`1` = device locks AWB and applies provided gains / 设备锁定 AWB 并应用所给增益 |
| `rgain` | `0.1` ~ `4.0` (manual only / 仅手动模式) | R channel gain. Higher `rgain` vs `bgain` shifts the image warm (yellow/red) / R 通道增益,相对 bgain 越大,画面越偏暖 |
| `bgain` | `0.1` ~ `4.0` (manual only / 仅手动模式) | B channel gain. Higher `bgain` vs `rgain` shifts the image cool (blue) / B 通道增益,相对 rgain 越大,画面越偏冷 |
> Gr / Gb channels are fixed to 1.0 by the device and are not adjustable.
> Gr / Gb 通道被设备固定为 1.0,不可调节。
#### 4.6.2 Response fields / 响应字段
All four services return a `success` (bool) and `rc` (int32). Get
services additionally return the queried state.
4 个 Service 都返回 `success` (bool) 与 `rc` (int32)。Get 类还会返回查询到的状态字段。
**`GetAe.Response`**
| Field | Typical range / 典型范围 | Meaning / 含义 |
|---|---|---|
| `exposure_time` | `0.0001`~`0.033` s | Current exposure / 当前曝光时间 |
| `gain` | `1.0`~`64.0` | Current analog gain / 当前模拟增益 |
| `iso` | `100`~`6400` | Equivalent ISO / 等效 ISO |
| `brightness` | `0`~`255` | Average frame brightness / 平均帧亮度 |
| `is_converged` | `0` or `1` | `1` = AE settled / AE 已收敛 |
| `env_lv` | `0`~`15` | Ambient luminance index, higher = brighter / 环境光强度指数,越大越亮 |
| `fps` | `~10` / `~14.5` / `~29` | Current frame rate / 当前帧率 |
**`GetAwb.Response`**
| Field | Typical range / 典型范围 | Meaning / 含义 |
|---|---|---|
| `rgain` / `bgain` | `0.1`~`4.0` | R / B channel gain / R / B 通道增益 |
| `grgain` / `gbgain` | `1.0` (fixed / 固定) | Gr / Gb gain, device-fixed / Gr / Gb 增益,设备固定 |
| `cct` | `2500`~`8000` K | Correlated color temperature / 相关色温 |
| `ccri` | `-50`~`50` | Color temp deviation index, 0 = on Planckian locus / 色温偏离指数,0 表示在普朗克轨迹上 |
| `is_converged` | `0` or `1` | `1` = AWB settled / AWB 已收敛 |
#### 4.6.3 `rc` return code / `rc` 返回码
| `rc` | Meaning / 含义 |
|---|---|
| `0` | Success / 成功 |
| `400` | Device payload too short / 设备载荷过短 |
| `401` | Device opcode not supported / 设备不支持该 opcode |
| `402` | Device parameter length wrong / 参数长度错误 |
| `403` | **Parameter out of range** / 参数越界 — most common when manual values exceed the table above / 手动值超出上表范围时最常见 |
| `404` | Device-side socket error / 设备端 socket 错误 |
| `405` | Device-side `ae_control` did not respond / 设备端 `ae_control` 无应答(确认 lydapp 已运行) |
| `255` (`0xFF`) | Unknown opcode reported by ae_control / ae_control 报未知 opcode |
| `-1` | SDK not initialised / SDK 未初始化 |
| `-2` ~ `-5` | USB transfer / timeout / malformed reply / USB 传输异常、超时、应答畸形 |
| `-100` | **Driver has not opened the device yet** / driver 还未打开设备,请等设备连接成功 |
#### 4.6.4 Usage examples / 调用示例
ROS2 (Humble) — start the driver in one terminal, then in a side terminal:
ROS2(Humble)—— 在一个终端启动 driver,在另一个终端:
```bash
source install/setup.bash
# Query current state / 查询当前状态
ros2 service call /odin1/get_ae odin_ros_driver/srv/GetAe
ros2 service call /odin1/get_awb odin_ros_driver/srv/GetAwb
# Set AE to AUTO / 设置 AE 为自动
ros2 service call /odin1/set_ae odin_ros_driver/srv/SetAe "{mode: 0}"
# Set AE to MANUAL with 10 ms exposure and gain 4.0
# 设置 AE 为手动,10 毫秒曝光,增益 4.0
ros2 service call /odin1/set_ae odin_ros_driver/srv/SetAe \
"{mode: 1, exposure_time: 0.010, gain: 4.0}"
# Set AWB to MANUAL with rgain=1.5, bgain=2.0
# 设置 AWB 为手动,rgain=1.5、bgain=2.0
ros2 service call /odin1/set_awb odin_ros_driver/srv/SetAwb \
"{mode: 1, rgain: 1.5, bgain: 2.0}"
# Restore AUTO / 一键回自动
ros2 service call /odin1/set_ae odin_ros_driver/srv/SetAe "{mode: 0}"
ros2 service call /odin1/set_awb odin_ros_driver/srv/SetAwb "{mode: 0}"
# Inspect srv definition / 查看 srv 完整定义
ros2 interface show odin_ros_driver/srv/SetAe
```
ROS1 (Noetic) — start the driver, then in a side terminal:
ROS1Noetic)—— 启动 driver 后,新开终端:
```bash
source devel/setup.bash
# Query / 查询
rosservice call /odin1/get_ae
rosservice call /odin1/get_awb
# Set AE manual / 设置 AE 手动
rosservice call /odin1/set_ae "{mode: 1, exposure_time: 0.010, gain: 4.0}"
# Set AWB manual / 设置 AWB 手动
rosservice call /odin1/set_awb "{mode: 1, rgain: 1.5, bgain: 2.0}"
# Restore AUTO (ROS1 requires all fields to be present)
# 一键回自动(ROS1 要求填齐全部字段)
rosservice call /odin1/set_ae "{mode: 0, exposure_time: 0.0, gain: 0.0}"
rosservice call /odin1/set_awb "{mode: 0, rgain: 0.0, bgain: 0.0}"
# Inspect srv definition / 查看 srv 完整定义
rossrv show odin_ros_driver/SetAe
```
#### 4.6.5 Recommended starting points by scene / 不同场景推荐起步参数
**AE (`exposure_time`, `gain`)**
| Scene / 场景 | `exposure_time` | `gain` |
|---|---|---|
| Bright outdoor / 明亮室外 | `0.001` ~ `0.005` s | `1.0` ~ `2.0` |
| Normal indoor / 普通室内 | `0.008` ~ `0.015` s | `2.0` ~ `8.0` |
| Dim light / 暗光环境 | `0.020` ~ `0.030` s | `8.0` ~ `32.0` |
| Very dark / 极暗 | `0.033` s | `32.0` ~ `64.0` |
**AWB (`rgain`, `bgain`)**
| Target tone / 目标色调 | `rgain` | `bgain` |
|---|---|---|
| Warm (tungsten, sunset) / 暖(钨丝灯、夕阳) | `2.0` ~ `2.5` | `1.0` ~ `1.2` |
| Neutral (D65 daylight) / 中性(D65 日光) | `1.5` ~ `1.7` | `1.8` ~ `2.0` |
| Cool (cloudy, fluorescent) / 冷(阴天、荧光) | `1.2` ~ `1.4` | `2.2` ~ `2.6` |
| Very cool / 极冷 | `1.0` | `3.0` ~ `4.0` |
#### 4.6.6 Caveats / 注意事项
- The service blocks for up to ~10 s waiting for the device to reply;
typical latency is tens of milliseconds.
Service 最长阻塞约 10 秒等设备应答;正常几十毫秒返回。
- Manual mode is **not** persisted across driver / device restart;
it falls back to AUTO on each new connection.
手动模式**不会**跨重启保留;每次重连默认回到 AUTO。
- `rc = -100` means the driver has not yet opened the device.
Wait until the driver logs `device connected` before calling.
返回 `rc = -100` 表示 driver 还没打开设备,等到 driver 日志显示 `device connected` 再调用。
- The effective maximum `exposure_time` is bounded by the frame
period `1 / fps`. With `dtof_fps = 290` (29 Hz, period ~34 ms)
the upper limit 0.033 s is already at the frame boundary.
最大可用 `exposure_time` 受帧周期 `1/fps` 限制。在 `dtof_fps = 290`29 Hz、周期 ~34 ms)下,上限 0.033 s 已经贴到帧边界。
## 5. FAQ
### 5.1 Segmentation fault upon re-launching host SDK
**Error Message**
No device connected after 60 seconds
**Solution**
1. Please power on Odin module again # Disconnect and reconnect odin power
2. Reinitialize Odin SDK # Execute SDK after device reboot
### 5.2 Library binding failure during compilation
**Error Message**
ld: cannot find -llydHostApi or symbol lookup errors
**Resolution**
1. Clean previous build artifacts
ROS1
```shell
rm -rf devel/ build/
```
ROS2
```shell
rm -rf devel/ install/ log/
```
2. Re-run script installation
### 5.3 Docker GUI passthrough failure
**Error Message**
Unable to open X display or No protocol specified
**Resolution**
```shell
xhost + #This command enables graphical passthrough to Docker containers
```
### 5.4 ROS driver exit with get version failed error
**Error Message**
```shell
<ERROR><api.cpp:lidar_get_version:672>: get device version fail.
get version failed.
```
**Resolution**
Device firmware version is too low, please update to latest version.
### 5.5 RVIZ has not responded for a long time
**Error Message**
Rviz does not respond, and after a while the terminal prints Device disconnected, waiting for reconnection...
**Resolution**
Please power on Odin module again
### 5.6 Device not responding
**Error Message**
Missed ok response from device,probably wrong interaction procedure.
**Resolution**
Please adopt the solution mentioned in 5.1
### 5.7 Device has no external calibration file
**Error Message**
ERRORMissing camera node 'cam_0'
**Resolution**
Please plug and unplug the USB again
### 5.8 ROS Driver report device disconnected immediately after stream started
**Error Message**
```shell
Device ready and streams activated
Device detaching...
Wating for device reconnection...
Device disconnected, waiting for reconnection...
```
**Reason**
Mostly common on ros2 environment and connected to complex network environment, such as office wifi & ethernet. ROS2 default to broadcast, and complex network environment will cause ros2 publish to block, leading to device disconnection.
**Resolution**
If cross-device communication is not required, please restrict ros2 to localhost only with:
```shell
export ROS_LOCALHOST_ONLY=1
```
If cross-device communication is required, please simplify the network environment as much as possible. Mini local network with only required devices is recommended.
### 5.9 ROS Driver died immediately after stream started
**Error Message**
```shell
Device ready and streams activated
[host_sdk_sample-2] process has died ......
```
**Test**
Disable odin1/image with sendrgb = 0 in control_command.yaml and try again. If the driver now works, it is likely that the issue is related to multiple version of opencv is installed on the system.
**Resolution**
Purge the unused version of opencv and maintain a single complete version, then rebuild the driver and try again.
### 5.10 ROS Driver printing "TF_OLD_DATA ignoring data" warning
**Error Message**
```shell
[rviz2-3] Warning: TF_OLD_DATA ignoring data from the past for frame odin1_base_link at time 20.547632 according to authority Authority undetectable
[rviz2-3] Possible reasons are listed at http://wiki.ros.org/tf/Errors%20explained
[rviz2-3] at line 294 in ./src/buffer_core.cpp
```
**Reason**
This is a ros & rviz feature to warn user that some tf data is being ignored due to timestamp conflicts. It happens when user keeps ros driver running and power-cycles odin device, which cause odin's internal system time being reset and now data timestamps conflicts with old data recieved by rviz during last run.
**Resolution**
There's a reset button on bottom of rviz gui. Click on this button will reset rviz's internal state and stop the warning.
### 5.11 ROS Driver printing "unknown cmd code: xx" error
**Error Message**
```shell
<ERROR><api.cpp:cmd_data_deal:418>: unknow command code 21.
```
**Reason**
This is due to ros driver version mismatch with device firmware version, resulting in ros driver unable to decode new data added in newer firmware.
**Resolution**
Please make sure you are using most up-to-date ros driver and device firmware.
### 5.12 USB device access error (LIBUSB_ERROR_BUSY or LIBUSB_ERROR_ACCESS)
**Error Message**
```shell
libusb: error [udev_hotplug_event] ignoring udev action bind
LIBUSB_ERROR_BUSY
```
or
```shell
libusb: error [_get_usbfs_fd] libusb couldn't open USB device /dev/bus/usb/xxx/xxx, errno=13
LIBUSB_ERROR_ACCESS
```
**Reason**
- **LIBUSB_ERROR_BUSY**: Another process is already using the USB device. This commonly happens when multiple instances of the ROS driver are running, or another application (such as a previous crashed instance) still holds the device handle.
- **LIBUSB_ERROR_ACCESS**: The current user does not have permission to access the USB device. This is typically caused by missing udev rules or insufficient user privileges.
**Resolution**
For **LIBUSB_ERROR_BUSY**:
1. Check if another instance of the driver is running:
```shell
ps aux | grep host_sdk_sample
```
2. Kill any existing instances:
```shell
killall host_sdk_sample
```
3. If the issue persists, unplug and replug the USB device to reset the device state.
For **LIBUSB_ERROR_ACCESS**:
1. Add udev rules for the device. Create a file `/etc/udev/rules.d/99-odin.rules` with the following content:
```shell
SUBSYSTEM=="usb", ATTR{idVendor}=="2207", ATTR{idProduct}=="0019", MODE="0666", GROUP="plugdev"
```
2. Reload udev rules:
```shell
sudo udevadm control --reload-rules
sudo udevadm trigger
```
3. Alternatively, run the driver with sudo (not recommended for production):
```shell
sudo -E ros2 launch odin_ros_driver odin_ros_driver.launch.py
```
4. Make sure your user is in the `plugdev` group:
```shell
sudo usermod -aG plugdev $USER
```
Then log out and log back in for the group change to take effect.
### 5.13 ros2 bag drops high-frequency topics (IMU / odometry_highfreq) / ros2 bag 录制丢失高频话题(IMU / odometry_highfreq
**Symptom / 现象**
When recording with `ros2 bag record`, low-frequency topics (cloud, image, odometry, wiwc) are intact, but `/odin1/imu` (400 Hz) and `/odin1/odometry_highfreq` (400 Hz) show missing samples — analysis scripts report inter-message intervals that are 2× or more of the expected period, while no drop is reported on the SDK side or by an online subscriber such as `ros2 topic hz`.
使用 `ros2 bag record` 录制时,低频话题(cloud、image、odometry、wiwc)完整无丢,但 `/odin1/imu`400 Hz)和 `/odin1/odometry_highfreq`(400 Hz)会出现丢帧——分析脚本上看到消息间隔达到正常周期的 2 倍以上,而 SDK 侧不报丢,独立的 `ros2 topic hz` 订阅者也看不到丢。
**Reason / 原因**
The driver publishes `/odin1/imu` and `/odin1/odometry_highfreq` with `RELIABLE` QoS. By default `ros2 bag record` subscribes with `history = keep_last`, `depth = 10`, which only buffers ~25 ms of samples at 400 Hz. Whenever the recorder is briefly delayed (disk flush, mcap/sqlite chunk write, scheduler jitter), its subscription queue overflows and DDS silently drops the oldest samples on the **subscriber side**. The SDK and publisher are unaffected, which is why no drop appears in the driver logs or in `ros2 topic hz`.
驱动以 `RELIABLE` QoS 发布 `/odin1/imu``/odin1/odometry_highfreq``ros2 bag record` 默认订阅使用 `history = keep_last``depth = 10`,在 400 Hz 下只能缓冲约 25 ms。一旦录制端有短暂阻塞(落盘 flush、mcap/sqlite chunk 写入、调度抖动),订阅队列就会溢出,DDS 在**订阅端**静默丢掉最旧的样本。SDK 与 publisher 不受影响,因此驱动日志和 `ros2 topic hz` 都看不到丢。
**Resolution / 解决方案**
Use the provided QoS override file `script/rosbag2_qos.yaml` to raise the subscriber-side queue depth on the recorder for the two high-rate topics:
使用本仓库提供的 QoS 配置 `script/rosbag2_qos.yaml`,把高频话题的录制订阅 depth 拉大:
```yaml
# script/rosbag2_qos.yaml
/odin1/imu:
reliability: reliable
history: keep_last
depth: 4000
/odin1/odometry_highfreq:
reliability: reliable
history: keep_last
depth: 4000
```
Apply it when recording / 录制时通过 `--qos-profile-overrides-path` 应用:
```shell
ros2 bag record -a \
--qos-profile-overrides-path src/odin_ros_driver/script/rosbag2_qos.yaml \
-o my_bag
```
Or only the high-rate topics / 也可以只录制高频话题:
```shell
ros2 bag record \
--qos-profile-overrides-path src/odin_ros_driver/script/rosbag2_qos.yaml \
-o my_bag \
/odin1/imu /odin1/odometry_highfreq /odin1/odometry /odin1/wiwc /odin1/cloud_raw
```
**Optional further tuning / 可选的进一步优化**
If drops still occur after applying the override (typically on slower disks), try the following in addition / 套用上述 override 后仍有丢包时(通常发生在慢盘上),可叠加以下措施:
```shell
# Use mcap backend with a larger internal cache (faster than sqlite3).
# 使用 mcap 后端 + 更大的内部缓存(比 sqlite3 快)。
ros2 bag record -s mcap --max-cache-size 1073741824 \
--qos-profile-overrides-path src/odin_ros_driver/script/rosbag2_qos.yaml \
-o my_bag \
/odin1/imu /odin1/odometry_highfreq ...
# Enlarge kernel UDP socket buffers (the most common hidden bottleneck for
# 400 Hz RELIABLE traffic, default is only 208 KB).
# 放大内核 UDP socket buffer400 Hz RELIABLE 流量最常见的隐藏瓶颈,默认仅 208 KB)。
sudo sysctl -w net.core.rmem_max=33554432
sudo sysctl -w net.core.wmem_max=33554432
```
**Does ROS1 have the same problem? / ROS1 是否存在同样的问题?**
No. ROS1 uses TCP-based publish/subscribe with a single `queue_size` parameter on each side, and has no QoS profile mismatch between publisher and subscriber. The ROS1 publisher path in this driver already sizes the IMU and `odometry_highfreq` publishers to `queue_size = 4000` (`include/host_sdk_sample.h`, see `initialize_publishers` ROS1 branch), and `rosbag record` uses TCP transport which is reliable by construction. As a result this specific drop pattern does not occur under ROS1; no additional configuration is required.
不存在。ROS1 使用基于 TCP 的发布/订阅,发布端与订阅端各自只有一个 `queue_size` 参数,不存在 ROS2 那种 QoS profile 不匹配的问题。本驱动 ROS1 路径已经把 IMU 与 `odometry_highfreq` 的发布队列设置为 `queue_size = 4000`(见 `include/host_sdk_sample.h``initialize_publishers` 的 ROS1 分支),并且 `rosbag record` 使用 TCP 传输本身即可靠传递。因此在 ROS1 下不会出现该丢帧现象,也不需要额外配置。
## 6. Contact Information
You can contact our support through support@manifoldtech.cn
To help diagnose the issue, please provide the following details to our FAE engineer:
1. Current firmware version
```shell
[device_version_capture]: ros_driver_version: [Version Number]
```
2. Photos of power adapter and converter cable in use.
3. Does the issue happen occasionally or consistently?
4. Provide images of the problem scenario.
5. Did the troubleshooting methods in Section V resolve the issue?
6. Expected timeline for issue resolution.
因此 `v0.10.0` 记录的是 ROS 2/C++ 迁移初版,不能仅凭本目录宣称 Odin 驱动可独立构建。兼容的 Odin ROS 2 驱动源码从后续版本开始随工程归档。
@@ -1,110 +0,0 @@
register_keys:
# if off, allow connection even if usb connection is below usb 3.0
# ATTENTION: usb 3.0 is always recommended, as advance functionality like SLAM mode requires usb 3.0 for reliable map file transfer
strict_usb3.0_check: 0 # 0: off: 1: on;
# 0: use odin internal system time as data time stamp, typical and recommended;
# 1: use host ros time (upon receive) as data time stamp, only use if you specifically require this setup, not recommended for most users
# 2: align odin1 time to host time, timestamp is the sensor data reception time on host time axis
use_host_ros_time: 1
streamctrl: 1 # 0: off; 1: on
# original rgb data in jpeg format from device
sendrgbcompressed: 1 # 0: off; 1: on
# RGB data, decoded from original jpeg data from device, bgr8 format
# Processed on host device
sendrgb: 1 # 0: off; 1: on
# undistort rgb image processed from decoded rgb data.
# depends on sendrgb. related camera parameters can be found in ws/src/odin_ros_driver/config/calib.yaml
# Processed on host device
sendrgbundistort: 0 # 0: off; 1: on.
# IMU data
sendimu: 1 # 0: off; 1: on
# SDK IMU smooth sending feature
# When enabled, SDK will send IMU data at precise intervals (default 400Hz)
# using a dedicated high-priority thread to reduce jitter and timing variance
enable_imu_smooth: 1 # 0: disable SDK IMU smooth sending; 1: enable (default)
# SDK IMU smooth sending frequency in Hz (only effective when enable_imu_smooth = 1)
imu_smooth_frequency: 400 # 1-1000 Hz, recommended 400 Hz
# Odometry data
sendodom: 1 # 0: off; 1: on
# TF from odom to base_link. Leave it on unless you specifically need it off.
# ATTENTION: critical for rviz to show cloud_raw.
send_odom_baselink_tf: 0 # 0: off; 1: on.
# raw dtof data
senddtof: 1 # 0: off; 1: on
cloud_raw_confidence_threshold: 35 # please refer to readme for more details
# dtof sensor frame rate. Supported values: 100 (10fps) or 145 (14.5fps)
# Higher frame rate provides smoother point cloud data but may increase data bandwidth
# Note: value is multiplied by 10 (e.g., 145 means 14.5fps)
dtof_fps: 100 # 100: 10fps; 145: 14.5fps
# slam cloud data
sendcloudslam: 1 # 0: off; 1: on
# processed with raw point cloud, rgb image, and calib.yaml from device
# Processed on host device
sendcloudrender: 1 # 0: off; 1: on
# depth completion demo, high computing resource usage.
# for more information please refer to the readme file.
# Processed on host device
senddepth: 0 # 0: off; 1: on
# cloud reprojection demo, projects cloud_slam to camera image using odometry
# Processed on host device
sendreprojection: 0 # 0: off; 1: on
# image overlay settings - overlays reprojected points on camera image
# Processed on host device
sendoverlay: 0 # 0: off; 1: on
overlay_reprojected_topic: "/odin1/reprojected_image" # reprojected image topic
overlay_camera_topic: "/odin1/image/undistorted" # camera image topic (undistorted)
overlay_output_topic: "/odin1/overlay_image" # output overlay image topic
overlay_alpha: 0.6 # blend alpha (0.0-1.0, higher = more reproj color)
# record rgb, odometry, and slam cloud data as proprietary olx format for further processing in MindCloud(TM) software.
# save path: ws/src/odin_ros_driver/recorddata/{record_start_time}/
# ATTENTION: please copy the full folder for post-processing.
recorddata: 0 # 0: off; 1: on
# Save device runtime status info to ws/src/odin_ros_driver/log/Driver_{drvier_start_time}/Conn_{device_connection_time}/dev_status.csv
devstatuslog: 1 # 0: off; 1: on.
save_log: 0 # 0: off; 1: on;
# raw dtof sensor intensity data in gray format, mostly for debug purpose.
pubintensitygray: 0 # 0: off; 1: on
showpath: 0 # 0: off; 1: on
showcamerapose: 0 # 0: off; 1: on
# Screen LOC=ODOM profile. Keep Odin in odometry mode; sim2real_web_udp_bridge_node
# may publish the pure-odom map->odom fallback anchor.
custom_map_mode: 0 # 0: Odometry mode 1: SLAM mode 2: Relocalization mode.
custom_init_pos: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]
custom_init_pose_search_radius: 4.0 # max position offset in meters, recommended <= 10
custom_init_pose_max_rot_deg: 180.0 # max rotation offset in degrees, up to 180
relocalization_map_abs_path: "" # unused in odometry mode
# To get the mapping result file, please use the set_param.sh script provided: "./set_param.sh save_map 1"
mapping_result_dest_dir: "" # "": if not specified, save to default location of {ws}/src/odin_ros_driver/map/{driver_start_time}/
mapping_result_file_name: "" # "": if not specified, save to location above with default file name of map_{map_save_time}.bin
# Image mask transfer settings
sendimagemask: 0 # 0: off; 1: on - transfer image mask to device on startup
image_mask_abs_path: "" # absolute path to the image mask file (e.g., /path/to/mask.png(1600x1296 resolution)
# Algorithm reset settings
resetalgo: 0 # 0: off; 1: on - send algo_reset command to device on startup
@@ -1,110 +0,0 @@
register_keys:
# if off, allow connection even if usb connection is below usb 3.0
# ATTENTION: usb 3.0 is always recommended, as advance functionality like SLAM mode requires usb 3.0 for reliable map file transfer
strict_usb3.0_check: 0 # 0: off: 1: on;
# 0: use odin internal system time as data time stamp, typical and recommended;
# 1: use host ros time (upon receive) as data time stamp, only use if you specifically require this setup, not recommended for most users
# 2: align odin1 time to host time, timestamp is the sensor data reception time on host time axis
use_host_ros_time: 1
streamctrl: 1 # 0: off; 1: on
# original rgb data in jpeg format from device
sendrgbcompressed: 1 # 0: off; 1: on
# RGB data, decoded from original jpeg data from device, bgr8 format
# Processed on host device
sendrgb: 1 # 0: off; 1: on
# undistort rgb image processed from decoded rgb data.
# depends on sendrgb. related camera parameters can be found in ws/src/odin_ros_driver/config/calib.yaml
# Processed on host device
sendrgbundistort: 0 # 0: off; 1: on.
# IMU data
sendimu: 1 # 0: off; 1: on
# SDK IMU smooth sending feature
# When enabled, SDK will send IMU data at precise intervals (default 400Hz)
# using a dedicated high-priority thread to reduce jitter and timing variance
enable_imu_smooth: 1 # 0: disable SDK IMU smooth sending; 1: enable (default)
# SDK IMU smooth sending frequency in Hz (only effective when enable_imu_smooth = 1)
imu_smooth_frequency: 400 # 1-1000 Hz, recommended 400 Hz
# Odometry data
sendodom: 1 # 0: off; 1: on
# TF from odom to base_link. Leave it on unless you specifically need it off.
# ATTENTION: critical for rviz to show cloud_raw.
send_odom_baselink_tf: 0 # 0: off; 1: on.
# raw dtof data
senddtof: 1 # 0: off; 1: on
cloud_raw_confidence_threshold: 35 # please refer to readme for more details
# dtof sensor frame rate. Supported values: 100 (10fps) or 145 (14.5fps)
# Higher frame rate provides smoother point cloud data but may increase data bandwidth
# Note: value is multiplied by 10 (e.g., 145 means 14.5fps)
dtof_fps: 100 # 100: 10fps; 145: 14.5fps
# slam cloud data
sendcloudslam: 1 # 0: off; 1: on
# processed with raw point cloud, rgb image, and calib.yaml from device
# Processed on host device
sendcloudrender: 1 # 0: off; 1: on
# depth completion demo, high computing resource usage.
# for more information please refer to the readme file.
# Processed on host device
senddepth: 0 # 0: off; 1: on
# cloud reprojection demo, projects cloud_slam to camera image using odometry
# Processed on host device
sendreprojection: 0 # 0: off; 1: on
# image overlay settings - overlays reprojected points on camera image
# Processed on host device
sendoverlay: 0 # 0: off; 1: on
overlay_reprojected_topic: "/odin1/reprojected_image" # reprojected image topic
overlay_camera_topic: "/odin1/image/undistorted" # camera image topic (undistorted)
overlay_output_topic: "/odin1/overlay_image" # output overlay image topic
overlay_alpha: 0.6 # blend alpha (0.0-1.0, higher = more reproj color)
# record rgb, odometry, and slam cloud data as proprietary olx format for further processing in MindCloud(TM) software.
# save path: ws/src/odin_ros_driver/recorddata/{record_start_time}/
# ATTENTION: please copy the full folder for post-processing.
recorddata: 0 # 0: off; 1: on
# Save device runtime status info to ws/src/odin_ros_driver/log/Driver_{drvier_start_time}/Conn_{device_connection_time}/dev_status.csv
devstatuslog: 1 # 0: off; 1: on.
save_log: 0 # 0: off; 1: on;
# raw dtof sensor intensity data in gray format, mostly for debug purpose.
pubintensitygray: 0 # 0: off; 1: on
showpath: 0 # 0: off; 1: on
showcamerapose: 0 # 0: off; 1: on
# Screen LOC=RELOC profile. Odin loads relocalization_map_abs_path and publishes
# the map/odom TF after relocalization succeeds; pure-odom fallback is disabled.
custom_map_mode: 2 # 0: Odometry mode 1: SLAM mode 2: Relocalization mode.
custom_init_pos: [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0]
custom_init_pose_search_radius: 4.0 # max position offset in meters, recommended <= 10
custom_init_pose_max_rot_deg: 180.0 # max rotation offset in degrees, up to 180
relocalization_map_abs_path: "/absolute/path/to/1hao.bin" # required for Relocalization mode; edit on the target computer
# To get the mapping result file, please use the set_param.sh script provided: "./set_param.sh save_map 1"
mapping_result_dest_dir: "" # "": if not specified, save to default location of {ws}/src/odin_ros_driver/map/{driver_start_time}/
mapping_result_file_name: "" # "": if not specified, save to location above with default file name of map_{map_save_time}.bin
# Image mask transfer settings
sendimagemask: 0 # 0: off; 1: on - transfer image mask to device on startup
image_mask_abs_path: "" # absolute path to the image mask file (e.g., /path/to/mask.png(1600x1296 resolution)
# Algorithm reset settings
resetalgo: 0 # 0: off; 1: on - send algo_reset command to device on startup
@@ -1,443 +0,0 @@
#ifndef LIDAR_TYPES_H
#define LIDAR_TYPES_H
#include <stdbool.h>
#include <stdlib.h>
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
#define LIDAR_SERIAL_MAX 64
#define LIDAR_MODEL_MAX 64
#define LIDAR_IP_MAX 64
typedef void * device_handle;
typedef enum {
LIDAR_LOG_ERROR = 0,
LIDAR_LOG_WARN,
LIDAR_LOG_INFO,
LIDAR_LOG_DEBUG,
} lidar_log_level_e;
typedef enum {
LIDAR_OTA_ALGORITHM,
LIDAR_OTA_FIRMWARE,
LIDAR_OTA_SCRIPT,
LIDAR_OTA_CALIBRATION
} lidar_ota_type_e;
typedef enum {
LIDAR_MODE_RAW,
LIDAR_MODE_SLAM,
} lidar_mode_e;
/**
* @brief Data types for lidar_data_callback_t
*
* Each type corresponds to a specific stream format in lidar_data_t.stream (capture_Image_List_t).
*
* ┌─────────────────────────────────────────────────────────────────────────────────────────────┐
* │ LIDAR_DT_RAW_RGB │
* │ imageCount: 1 │
* │ imageList[0]: NV12 image data │
* │ - pAddr: uint8_t* (Y plane followed by UV plane) │
* │ - width: 1536, height: 1280 │
* │ - length: width * height * 3 / 2 bytes │
* ├─────────────────────────────────────────────────────────────────────────────────────────────┤
* │ LIDAR_DT_RAW_IMU │
* │ imageCount: 1 │
* │ imageList[0]: IMU data │
* │ - pAddr: imu_convert_data_t* │
* │ - accel[3]: float (m/s^2) │
* │ - gyro[3]: float (rad/s) │
* │ - stamp: uint64_t (ns) │
* │ - sequence: uint64_t │
* │ - length: sizeof(imu_convert_data_t) │
* ├─────────────────────────────────────────────────────────────────────────────────────────────┤
* │ LIDAR_DT_RAW_DTOF │
* │ imageCount: 4 │
* │ Resolution: 256 x 192 │
* │ imageList[0]: Depth image │
* │ - pAddr: float* (depth in meters) │
* │ - length: 256 * 192 * sizeof(float) │
* │ imageList[1]: Point cloud XYZ │
* │ - pAddr: float* (x,y,z interleaved) │
* │ - length: 256 * 192 * 3 * sizeof(float) │
* │ imageList[2]: Confidence │
* │ - pAddr: uint8_t* │
* │ - length: 256 * 192 * sizeof(uint8_t) │
* │ imageList[3]: Intensity/Reflectivity │
* │ - pAddr: uint16_t* │
* │ - length: 256 * 192 * sizeof(uint16_t) │
* ├─────────────────────────────────────────────────────────────────────────────────────────────┤
* │ LIDAR_DT_SLAM_CLOUD │
* │ imageCount: 1 │
* │ imageList[0]: SLAM point cloud (XYZRGBA, fixed-point on the wire) │
* │ - pAddr: slam_cloud_point_t* (7 * int32_t per point, see struct below) │
* │ - length: num_points * sizeof(slam_cloud_point_t) ( == num_points * 28 bytes ) │
* │ xyz are stored in 0.1 mm units: meters = xyz * SLAM_CLOUD_XYZ_TO_M │
* ├─────────────────────────────────────────────────────────────────────────────────────────────┤
* │ LIDAR_DT_SLAM_ODOMETRY / LIDAR_DT_SLAM_ODOMETRY_HIGHFREQ / LIDAR_DT_SLAM_ODOMETRY_TF │
* │ imageCount: 1 │
* │ imageList[0]: Odometry data │
* │ - pAddr: ros_odom_convert_complete_t* │
* │ - timestamp_ns: uint64_t │
* │ - pos[3]: int64_t (x,y,z in μm, divide by 1e6 for meters) │
* │ - orient[4]: int64_t (quaternion x,y,z,w, divide by 1e6) │
* │ - linear_velocity[3]: int64_t │
* │ - angular_velocity[3]: int64_t │
* │ - pose_cov[36]: double │
* │ - twist_cov[36]: double │
* │ - length: sizeof(ros_odom_convert_complete_t) │
* ├─────────────────────────────────────────────────────────────────────────────────────────────┤
* │ LIDAR_DT_DEV_STATUS │
* │ imageCount: 1 │
* │ imageList[0]: Device status │
* │ - pAddr: lidar_device_status_t* │
* │ - length: sizeof(lidar_device_status_t) │
* ├─────────────────────────────────────────────────────────────────────────────────────────────┤
* │ LIDAR_DT_NTP │
* │ imageCount: 1 │
* │ imageList[0]: PTP/NTP sync data │
* │ - pAddr: ptp_sync_data_t* │
* │ - delay: double │
* │ - offset: double │
* │ - length: sizeof(ptp_sync_data_t) │
* └─────────────────────────────────────────────────────────────────────────────────────────────┘
*/
typedef enum {
LIDAR_DT_NONE = 0, /**< No data */
LIDAR_DT_RAW_RGB, /**< RGB image (NV12 format, 1536x1280) */
LIDAR_DT_RAW_IMU, /**< IMU data (imu_convert_data_t) */
LIDAR_DT_RAW_DTOF, /**< DTOF raw data (depth + xyz + confidence + intensity, 256x192) */
LIDAR_DT_SLAM_CLOUD, /**< SLAM point cloud (XYZRGBA) */
LIDAR_DT_SLAM_ODOMETRY, /**< SLAM odometry (ros_odom_convert_complete_t) */
LIDAR_DT_DEV_STATUS, /**< Device status (lidar_device_status_t) */
LIDAR_DT_SLAM_ODOMETRY_HIGHFREQ,/**< High frequency odometry (ros_odom_convert_complete_t) */
LIDAR_DT_SLAM_ODOMETRY_TF, /**< Map-Odom TF transform (ros_odom_convert_complete_t) */
LIDAR_DT_SLAM_WIWC, /**< WIWC odometry */
LIDAR_DT_NTP /**< PTP/NTP sync data (ptp_sync_data_t) */
} lidar_data_type_e;
typedef struct {
int8_t serial[LIDAR_SERIAL_MAX];
int8_t model[LIDAR_MODEL_MAX];
bool online;
uint32_t initial_state;
} lidar_device_info_t;
typedef struct {
float x, y, z;
float intensity;
} lidar_point_t;
typedef struct {
float intrinsics[9];
float extrinsics[16];
} lidar_calibration_t;
#define DEVICE_MAX_CH_NUMBER 4
typedef struct {
uint64_t timestamp_ns;
int64_t pos[3];
int64_t orient[4];
} ros2_odom_convert_t;
typedef struct {
uint64_t timestamp_ns;
int64_t pos[3]; // x, y, z in μm
int64_t orient[4]; // quaternion x, y, z, w in 1e6 precision
int64_t linear_velocity[3];
int64_t angular_velocity[3];
double pose_cov[36];
double twist_cov[36];
} ros_odom_convert_complete_t;
typedef struct {
double delay;
double offset;
} ptp_sync_data_t;
/* ---------------------------------------------------------------------
* SLAM cloud wire format.
*
* One LIDAR_DT_SLAM_CLOUD point on the bus is a 7 * int32_t record:
* xyz[0..2] : x, y, z in 0.1 mm fixed-point.
* meters = xyz * SLAM_CLOUD_XYZ_TO_M.
* rgba[0..3]: r, g, b, a; each stored in the low byte of an int32_t.
*
* Consumers (SDK hooks, ROS driver, etc.) should reference this struct
* and the scale macro below as the single source of truth rather than
* re-hardcoding the stride or the divisor.
* ------------------------------------------------------------------- */
#define SLAM_CLOUD_XYZ_TO_M (1.0e-4) /* device 0.1mm units -> meters */
#define SLAM_CLOUD_XYZ_FROM_M (1.0e4) /* meters -> device 0.1mm units */
typedef struct {
int32_t xyz[3]; /* x, y, z in 0.1 mm fixed-point */
int32_t rgba[4]; /* r, g, b, a; only low byte of each is meaningful */
} slam_cloud_point_t;
typedef struct icm_6aixs_data_t {
int16_t aacx;
int16_t aacy;
int16_t aacz;
int16_t gyrox;
int16_t gyroy;
int16_t gyroz;
uint8_t valid;
uint32_t nums;
uint8_t fsync_pack;
uint16_t interval;
uint64_t stamp;
} icm_6aixs_data_t;
typedef struct {
float accel_x;
float accel_y;
float accel_z;
float gyro_x;
float gyro_y;
float gyro_z;
uint64_t stamp;
uint64_t sequence;
} imu_convert_data_t;
typedef struct {
uint32_t length;
uint64_t sequence;
uint64_t timestamp;
uint64_t interval;
void* pAddr;
uint32_t width;
uint32_t height;
} buffer_List_t;
typedef struct capture_Image_List_t {
uint32_t imageCount;
buffer_List_t imageList[DEVICE_MAX_CH_NUMBER];
} capture_Image_List_t;
typedef struct {
uint32_t type;
capture_Image_List_t stream;
} lidar_data_t;
typedef void (*lidar_device_callback_t)(const lidar_device_info_t* device, bool attach);
typedef void (*lidar_data_callback_t)(const lidar_data_t *data, void *user_data);
typedef struct {
lidar_data_callback_t data_callback;
void *user_data;
} lidar_data_callback_info_t;
typedef struct {
int major;
int minor;
int patch;
}lidar_version_t;
typedef struct {
lidar_version_t kernel_version;
lidar_version_t mcu_version;
lidar_version_t soc_version;
lidar_version_t Daemon_proc_version;
lidar_version_t slam_version;
} lidar_fireware_version_t;
/**
* @brief RGB image sensor frame rate
*
*/
typedef struct{
int configured_odr;/*rgb image sensor frame rate, offset: */
int tx_odr;/*The actual rgb image sensor frame rate, offset: */
} lidar_rgb_sensor_status_t;
/**
* @brief DTOF Lidar frame rate
*
*/
typedef struct{
int configured_odr;/* dtof lidar sensor frame rate, offset: */
int tx_odr;/*The actual dtof lidar sensor frame rate, offset: */
int subframe_odr;/*DTOF 6行为一组 这个是组间隔时间*/
short tx_temp;/* dtof lidar tx temp offset: */
short rx_temp;/* dtof lidar rx temp offset: */
} lidar_dtof_sensor_status_t;
/**
* @brief IMU Sensor
*
*/
typedef struct{
int configured_odr;
int tx_odr;
} lidar_imu_sensor_status_t;
typedef struct{
int package_temp;/*SOC整体温度*/
// int bigcore_temp;/*大核集群温度:4*A76*/
// int littlecore_temp;/*小核集群温度:4*A53*/
int cpu_temp;
int center_temp;/*SOC中心温度:4*A53*/
int gpu_temp;/* GPU模块温度 */
int npu_temp;/* NPU模块温度 */
} lidar_soc_thermal_t;
typedef struct
{
double uptime_seconds;
lidar_soc_thermal_t soc_thermal; /*offset: 0*/
int cpu_use_rate[8];/*cpu 使用率,offset: */
int ram_use_rate;/*运行内存使用率 ,offset: */
lidar_rgb_sensor_status_t rgb_sensor;
lidar_dtof_sensor_status_t dtof_sensor;
lidar_imu_sensor_status_t imu_sensor;
int slam_cloud_tx_odr; /* Actual frame rate of slam cloud offset: */
int slam_odom_tx_odr; /* Actual frame rate of slam odom offset: */
int slam_odom_highfreq_tx_odr; /* Actual frame rate of slam odom offset: */
} lidar_device_status_t;
typedef enum {
LIDAR_DEVICE_NONE = 0,
LIDAR_DEVICE_NOT_INITIALIZED,
LIDAR_DEVICE_INITIALIZED,
LIDAR_DEVICE_STREAMING,
LIDAR_DEVICE_STREAM_STOPPED,
} lidar_device_initial_state_e;
typedef enum {
LIDAR_DEPTH_ODR_10HZ = 0,
LIDAR_DEPTH_ODR_14_5HZ = 1,
LIDAR_DEPTH_ODR_29HZ = 2,
} lidar_depth_odr_e;
typedef struct {
lidar_depth_odr_e odr;
} lidar_depth_para_t;
/* ---------------------------------------------------------------------
* Camera AE / AWB control types.
*
* The host SDK forwards AE/AWB requests through the USB control channel
* (CMD_CODE_CONTROL_CMD + SYS_CONTROL_AE_UDP). The device-side lydapp
* relays them via UDP loopback to its ISP service. See
* sdk/api/Host_USB_AE_Protocol.md for the wire-level details.
* ------------------------------------------------------------------- */
/**
* @brief AE/AWB control mode used by lidar_set_ae_param / lidar_set_awb_param.
*
* - AUTO : the device runs its own AE/AWB convergence loop. The two
* float parameters of the corresponding Set call are ignored.
* - MANUAL : the device locks AE/AWB and applies the user-supplied
* (exposure_time, gain) or (rgain, bgain). Out-of-range
* values are rejected with rc = 403
* (LIDAR_AE_PARAM_OUT_OF_RANGE).
*/
typedef enum {
LIDAR_CAM_MODE_AUTO = 0, /**< switch to auto AE / AWB */
LIDAR_CAM_MODE_MANUAL = 1, /**< switch to manual AE / AWB and apply params */
} lidar_cam_mode_e;
/**
* @brief Current AE status returned by lidar_get_ae_info().
*
* Field-by-field meaning and typical range:
*
* exposure_time : current sensor exposure time, in seconds.
* Manual-mode valid range: 0.0001 .. 0.033.
* In auto mode varies with scene illumination.
* gain : current analog gain (linear, not dB).
* Manual-mode valid range: 1.0 .. 64.0.
* Higher value = brighter output but worse SNR.
* iso : equivalent ISO speed, typically 100 .. 6400.
* Derived from gain; informational only.
* brightness : average frame brightness in [0, 255]. AE target
* converges toward a mid-range value.
* is_converged : 1 = AE has settled, 0 = still adjusting.
* env_lv : ambient luminance index, typically 0 .. 15
* (higher = brighter scene).
* fps : actual frame rate in Hz, follows dtof_fps config
* (~10 / 14.5 / 29).
*/
typedef struct {
float exposure_time; /**< current exposure time (s), 0.0001..0.033 */
float gain; /**< current analog gain, 1.0..64.0 */
int32_t iso; /**< equivalent ISO, ~100..6400 */
float brightness; /**< average frame brightness, 0..255 */
uint8_t is_converged; /**< 1 = AE converged, 0 = not converged */
float env_lv; /**< ambient luminance level, ~0..15 */
float fps; /**< current frame rate (Hz) */
} lidar_ae_info_t;
/**
* @brief Current AWB status returned by lidar_get_awb_info().
*
* Field-by-field meaning and typical range:
*
* rgain : R channel gain. Manual-mode valid range: 0.1 .. 4.0.
* Raising rgain relative to bgain shifts the image
* toward warm (yellow/red).
* grgain : Gr channel gain. Device-fixed at 1.0, not adjustable.
* gbgain : Gb channel gain. Device-fixed at 1.0, not adjustable.
* bgain : B channel gain. Manual-mode valid range: 0.1 .. 4.0.
* Raising bgain relative to rgain shifts the image
* toward cool (blue).
* cct : correlated color temperature in Kelvin, typically
* 2500 .. 8000 K.
* ccri : color temperature deviation index, signed value
* roughly in -50 .. 50; 0 = on the Planckian locus.
* is_converged : 1 = AWB has settled, 0 = still adjusting.
*/
typedef struct {
float rgain; /**< R channel gain, 0.1..4.0 */
float grgain; /**< Gr channel gain, device-fixed 1.0 */
float gbgain; /**< Gb channel gain, device-fixed 1.0 */
float bgain; /**< B channel gain, 0.1..4.0 */
float cct; /**< color temperature (K), ~2500..8000 */
float ccri; /**< color temperature deviation, ~-50..50 */
uint8_t is_converged; /**< 1 = AWB converged, 0 = not converged */
} lidar_awb_info_t;
/**
* @brief AE/AWB device-side error codes.
*
* Mapped to positive return values of lidar_get_ae_info / lidar_set_ae_param /
* lidar_get_awb_info / lidar_set_awb_param when the device replies with
* CMD_CODE_FAIL. See Host_USB_AE_Protocol.md section 6.
*/
typedef enum {
LIDAR_AE_OK = 0,
LIDAR_AE_BAD_REQUEST = 400, /**< payload too short */
LIDAR_AE_UNSUPPORTED_OPCODE = 401,
LIDAR_AE_BAD_PARAM_LEN = 402,
LIDAR_AE_PARAM_OUT_OF_RANGE = 403,
LIDAR_AE_SOCKET_ERROR = 404,
LIDAR_AE_NO_RESPONSE = 405, /**< ae_control UDP timeout */
LIDAR_AE_UNKNOWN_OPCODE = 0xFF,/**< status byte from ae_control */
} lidar_ae_error_e;
#ifdef __cplusplus
}
#endif
#endif
@@ -1,18 +0,0 @@
# rosbag2_qos.yaml
# QoS overrides for rosbag2 recording
# 用于 rosbag2 录制的 QoS 配置
#
# Usage / 使用方法:
# ros2 bag record -a --qos-profile-overrides-path rosbag2_qos.yaml
# High frequency data (IMU 400Hz, Odom 100Hz)
# 高频数据 (IMU 400Hz, 里程计 100Hz)
/odin1/imu:
reliability: reliable
history: keep_last
depth: 4000
/odin1/odometry_highfreq:
reliability: reliable
history: keep_last
depth: 4000
@@ -1,13 +0,0 @@
# Query current AE (auto exposure) state.
# Request: empty.
# Response: success + raw return code + decoded fields.
---
bool success # true if rc == 0
int32 rc # 0=ok, >0=device error (400..405 or 0xFF), <0=SDK error
float32 exposure_time # seconds (manual range 0.0001 .. 0.033)
float32 gain # analog gain
int32 iso # equivalent ISO
float32 brightness # average frame brightness
uint8 is_converged # 1=converged, 0=not converged
float32 env_lv # ambient luminance level
float32 fps # current frame rate
@@ -1,13 +0,0 @@
# Query current AWB (auto white balance) state.
# Request: empty.
# Response: success + raw return code + decoded fields.
---
bool success # true if rc == 0
int32 rc # 0=ok, >0=device error (400..405 or 0xFF), <0=SDK error
float32 rgain
float32 grgain
float32 gbgain
float32 bgain
float32 cct # color temperature in Kelvin
float32 ccri # color temperature deviation
uint8 is_converged # 1=converged, 0=not converged
@@ -1,14 +0,0 @@
# Set AE (auto exposure) mode and (in manual mode) exposure / gain.
#
# mode == 0 (AUTO) : sends opcode 0x02 only; exposure_time and gain ignored.
# mode == 1 (MANUAL) : sends opcode 0x03 then 0x06 (exposure_time, gain).
#
# Valid manual ranges (device-enforced; out-of-range returns rc=403):
# exposure_time : 0.0001 s .. 0.033 s
# gain : 1.0 .. 64.0
uint8 mode
float32 exposure_time
float32 gain
---
bool success
int32 rc # 0=ok, >0=device error (400..405 or 0xFF), <0=SDK error
@@ -1,15 +0,0 @@
# Set AWB (auto white balance) mode and (in manual mode) R/B gains.
#
# mode == 0 (AUTO) : sends opcode 0x31 only; rgain and bgain ignored.
# mode == 1 (MANUAL) : sends opcode 0x32 then 0x33 (rgain, bgain).
# Gr/Gb are fixed to 1.0 by the device.
#
# Valid manual ranges:
# rgain : 0.1 .. 4.0
# bgain : 0.1 .. 4.0
uint8 mode
float32 rgain
float32 bgain
---
bool success
int32 rc # 0=ok, >0=device error (400..405 or 0xFF), <0=SDK error
@@ -47,7 +47,7 @@ action:
- rr_wheel
wheel_indices: [12, 13, 14, 15]
scale: [0.125, 0.25, 0.25, 0.125, 0.25, 0.25, 0.125, 0.25, 0.25, 0.125, 0.25, 0.25, 5.0, 5.0, 5.0, 5.0]
default_dof_pos: [0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.0, 0.0, 0.0]
default_dof_pos: [0.0, 0.9, -1.8, 0.0, 0.9, -1.8, 0.0, 0.9, -1.8, 0.0, 0.9, -1.8, 0.0, 0.0, 0.0, 0.0]
motor_mapping:
can_id_map:
@@ -4,47 +4,8 @@
motor_hz: 200.0
status_hz: 10.0
target_timeout_ms: 150.0
event_log_dir: "logs_v2_web"
model_engine_path: policies/model_6800_fp16.engine
prefer_tensorrt: true
model_path: policies/model_6800.onnx
rough_model_engine_path: policies/model_6800_fp16.engine
crawl_model_path: policies/model_crawl.onnx # unused while crawl_backend is "ik"
crawl_model_engine_path: ""
wall_model_path: policies/model_84.onnx
wall_model_engine_path: policies/model_84_fp16.engine
rough_default_dof_pos: [0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.0, 0.0, 0.0]
wall_default_dof_pos: [0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.550, -1.125, 0.0, 0.0, 0.0, 0.0]
crawl_backend: "ik"
crawl_default_dof_pos: [0.2, 1.697, -2.650, -0.2, 1.697, -2.650, 0.2, 1.697, -2.650, -0.2, 1.697, -2.650, 0.0, 0.0, 0.0, 0.0]
crawl_ik_wheel_linear_gain: 12.5
crawl_ik_wheel_yaw_gain: 8.0
crawl_ik_max_wheel_speed: 12.0
crawl_ik_abduction_clip: 0.45
crawl_ik_yaw_rate_kp: 0.5
crawl_ik_imu_posture: false
crawl_ik_encoder_posture_kp: 0.0
crawl_ik_encoder_posture_max: 0.03
crawl_ik_encoder_guard: false
crawl_ik_encoder_guard_start: 0.28
crawl_ik_encoder_guard_stop: 0.65
crawl_ik_imu_guard: true
crawl_ik_imu_guard_start_deg: 12.0
crawl_ik_imu_guard_stop_deg: 28.0
model_switch_transition_s: 0.9
model_switch_min_transition_s: 0.4
model_switch_to_stand_transition_scale: 2.1
model_switch_to_model_transition_scale: 2.4
model_switch_stand_hold_s: 0.45
model_switch_stand_max_err: 0.18
model_switch_stand_max_vel: 0.8
model_switch_release_scale: 1.0
runtime_max_vx: 0.9
runtime_max_vy: 0.5
runtime_max_yaw_rate: 0.85
debug_trace_enabled: true
debug_trace_decimation: 1
use_cuda: true # enable CUDA Execution Provider on Orin Nano GPU
model_path: policies/model_rough.onnx
use_cuda: true # 启用 CUDA Execution ProviderOrin Nano GPU 加速)
contract_file: deployment_contract.yaml
dry_run: false
can0_name: "can0"
@@ -52,60 +13,39 @@
imu_topic: "/odin1/imu"
odom_topic: "/odom"
# Remote UART / SBUS parameters, aligned with the first-generation Python deployment
remote_enabled: false # true
# Remote UART / SBUS parameters, aligned with the Python deployment
remote_enabled: true
remote_port: "/dev/ttyACM0"
remote_baudrate: 100000
remote_timeout: 0.02
remote_axis_deadzone: 40
remote_active_threshold: 40
remote_axis_full_scale: 660.0
remote_max_vx: 0.9
remote_max_vy: 0.5
remote_max_yaw_rate: 0.85
remote_max_vx: 0.8
remote_max_vy: 0.3
remote_max_yaw_rate: 0.5
remote_invert_vx: true
remote_invert_vy: false
remote_invert_yaw: true
remote_publish_inactive_zero: true
remote_estop_latch: true
remote_poll_hz: 50.0
remote_model_switch_enabled: true
remote_model_switch_channel: 10
remote_model_switch_debounce_frames: 3
remote_model_switch_rough_level: "low"
remote_model_switch_ik_level: "high"
# Command mux parameters
cmd_mux_default_mode: "NAV"
cmd_mux_default_mode: "REMOTE"
cmd_mux_output_hz: 50.0
cmd_mux_remote_timeout_ms: 250.0
cmd_mux_web_timeout_ms: 300.0
cmd_mux_nav_timeout_ms: 500.0
cmd_mux_max_vx: 0.9
cmd_mux_max_vy: 0.5
cmd_mux_max_yaw_rate: 0.85
cmd_mux_max_vx: 0.8
cmd_mux_max_vy: 0.3
cmd_mux_max_yaw_rate: 0.5
cmd_mux_max_vx_acc: 1.0
cmd_mux_max_vy_acc: 1.0
cmd_mux_max_yaw_acc: 1.5
cmd_mux_max_vx_decel: 2.0
cmd_mux_max_vy_decel: 2.0
cmd_mux_max_yaw_decel: 2.0
# The rough locomotion policy has an approximately 0.2 m/s linear command dead zone.
# Skip that ineffective band on start-up, but still allow exact zero for braking/estop.
cmd_mux_linear_deadzone_epsilon: 0.05
cmd_mux_yaw_deadzone_epsilon: 0.02
cmd_mux_min_effective_vx: 0.22
cmd_mux_min_effective_vy: 0.22
cmd_mux_min_effective_yaw_rate: 0.0
cmd_mux_deadzone_sources: "nav"
# Windows/Nano Web UDP bridge parameters
web_bridge_enabled: true
web_http_host: "0.0.0.0"
web_http_port: 18080
web_static_dir: ""
# Odom task actual path export; files can be opened by nav_tools over the PCD.
odom_trace_export_dir: "map/load"
web_udp_listen_host: "0.0.0.0"
web_udp_listen_port: 15000
web_udp_remote_host: ""
@@ -113,46 +53,22 @@
web_udp_state_hz: 20.0
web_udp_cmd_timeout_ms: 300.0
web_udp_max_packet_bytes: 8192
web_udp_max_vx: 0.9
web_udp_max_vx: 0.8
web_udp_max_vy: 0.3
web_udp_max_yaw_rate: 0.85
web_udp_max_yaw_rate: 0.5
web_udp_estop_on_timeout: false
# Safety parameters
safety_enabled: true
max_target_offset: 2.4
model_switch_max_target_offset: 1.8
hard_target_offset: 3.0
max_ang_vel: 30.0
max_target_offset: 0.6
hard_target_offset: 2.0
max_ang_vel: 10.0
max_tilt_z: -0.3
clip_to_brake: 0
imu_age_warn_ms: 60.0
imu_age_stop_ms: 500.0
wheel_no_effect_command_threshold: 1.0
wheel_no_effect_min_response_ratio: 0.20
wheel_no_effect_velocity_epsilon: 0.25
wheel_no_effect_max_temperature_c: 90.0
wheel_no_effect_min_bus_voltage_v: 18.0
wheel_no_effect_command_warmup_cycles: 12
wheel_no_effect_trigger_cycles: 30
wheel_no_effect_attempt_limit: 2
wheel_no_effect_cooldown_ms: 1200
wheel_recovery_verify_timeout_ms: 180
wheel_no_effect_diag_freshness_ms: 350
wheel_no_effect_diag_request_period_ms: 80
leg_no_effect_position_error_threshold: 0.18
leg_no_effect_velocity_epsilon: 0.12
leg_no_effect_max_estimated_current_arms: 4.0
leg_no_effect_max_abs_torque_nm: 5.0
leg_no_effect_max_temperature_c: 100.0
leg_no_effect_min_bus_voltage_v: 18.0
leg_no_effect_command_warmup_cycles: 40
leg_no_effect_trigger_cycles: 25
leg_no_effect_attempt_limit: 2
leg_no_effect_cooldown_ms: 1200
leg_recovery_verify_timeout_ms: 220
imu_age_stop_ms: 200.0
# Policy alignment with the first-generation Python deployment
# Policy alignment with the Python deployment
command_release_s: 0.35
release_command_hold_s: 0.12
release_posture_max_err: 0.35
@@ -162,143 +78,3 @@
enable_zero_cmd_suppression: true
require_active_command_to_release: true
zero_cmd_use_yaw_rate: true
# Simple navigation parameters
localization_mode: "relocal" # relocal: wait for Odin map/odom TF; odom: bridge map->odom fallback
nav_map_frame: "map"
nav_odom_frame: "odom"
nav_base_frame: "base_link"
nav_control_hz: 20.0
nav_goal_tolerance: 0.20
nav_yaw_stop_threshold: 0.80
nav_max_vx: 0.90
nav_max_vy: 0.50
nav_max_wz: 0.85
nav_kp_dist: 0.80
nav_kp_yaw: 1.80
nav_goal_exit_tolerance_margin: 0.08
nav_goal_complete_stable_cycles: 2
nav_final_align_kp_yaw_scale: 0.60
nav_final_align_max_wz: 0.45
nav_final_align_creep_speed: 0.05
nav_goal_yaw_tolerance_deg: 12.0
nav_astar_enabled: true
nav_astar_resolution: 0.10
nav_astar_pcd_sample_step: 5
nav_astar_allow_diagonal: true
nav_astar_smooth_enabled: true
nav_astar_corner_blend_dist: 0.20
nav_astar_waypoint_reach_dist: 0.18
nav_astar_lookahead_dist: 0.35
nav_astar_snap_radius: 0.60
nav_astar_max_expansions: 120000
# Slalom is treated as a continuous path by simple_nav even if the route JSON
# was saved without precisionFollow/stableCycles/lookahead metadata.
nav_slalom_auto_precision_enabled: true
nav_slalom_auto_precision_force: true
nav_slalom_task_names: "slalom"
nav_slalom_stable_cycles: 0
nav_slalom_lookahead: 0.35
nav_slalom_yaw_rate_limit: 0.45
nav_slalom_tolerance: 0.15
nav_slalom_max_vx: 0.58
nav_slalom_min_vx: 0.22
nav_slalom_curvature_slowdown_enabled: true
nav_slalom_min_turn_speed_scale: 0.45
# Execute waypoints marked slalomStraight as odometry-closed scripted moves.
nav_slalom_script_enabled: true
nav_slalom_script_start_tolerance: 0.22
nav_slalom_script_pos_tolerance: 0.10
nav_slalom_script_yaw_tolerance_deg: 5.0
nav_slalom_script_drive_yaw_deadband_deg: 8.0
nav_slalom_script_stable_cycles: 1
nav_slalom_script_rotate_steps_enabled: false
nav_slalom_script_final_rotate_enabled: false
nav_slalom_script_require_yaw_at_step: false
nav_slalom_script_kp_dist: 1.00
nav_slalom_script_kp_yaw: 1.20
nav_slalom_script_max_vx: 0.58
nav_slalom_script_max_vy: 0.50
nav_slalom_script_max_wz: 0.50
nav_slalom_script_min_cmd_linear: 0.22
nav_slalom_script_min_cmd_angular: 0.20
nav_slalom_script_min_cmd_epsilon: 0.05
nav_slalom_script_min_step_distance: 0.02
nav_slalom_script_yaw_gate_deg: 8.0
nav_slalom_script_lateral_gate: 0.07
nav_slalom_script_lateral_slow_gate: 0.15
nav_slalom_script_lateral_creep_vx: 0.22
nav_slalom_script_drive_yaw_source: "segment"
nav_slalom_script_segment_yaw_min_dist: 0.45
nav_precision_lateral_control_enabled: true
nav_precision_lateral_kp: 0.80
nav_precision_lateral_max_vy: 0.22
# Lightweight DWA-style local safety layer over the route/avoid polygons.
nav_local_planner_enabled: true
nav_local_planner_tasks: "slalom"
nav_local_planner_precision_enabled: true
nav_local_planner_sim_time: 0.9
nav_local_planner_sim_dt: 0.1
nav_local_planner_v_samples: 5
nav_local_planner_w_samples: 7
nav_local_planner_vy_samples: 3
nav_local_planner_obstacle_margin: 0.08
nav_local_planner_recovery_clearance_epsilon: 0.005
# 0.0 means auto: use the nav_tools body+wheel lateral footprint.
nav_local_planner_robot_radius: 0.0
nav_local_planner_clearance_weight: 2.0
nav_local_planner_path_weight: 2.0
nav_local_planner_heading_weight: 0.7
nav_local_planner_speed_weight: 0.3
nav_local_planner_nominal_weight: 1.0
nav_local_planner_min_vx: 0.22
nav_slalom_script_safety_filter_enabled: true
nav_local_planner_use_astar_grid: false
nav_turn_in_place_enabled: true
nav_turn_in_place_enter_yaw_deg: 70.0
nav_turn_in_place_exit_yaw_deg: 18.0
nav_turn_in_place_max_wz: 0.80
nav_pre_dock_enabled: true
nav_pre_dock_distance: 0.35
nav_pre_dock_tolerance: 0.18
nav_pre_dock_skip_within_goal_dist: 0.45
nav_goals_file: ""
nav_missions_file: ""
# Keep the legacy YAML route for reference; active task is selected by nav_route_task_file below.
nav_route_file: ""
# Task switch entry: change this path to another .json/.yaml route file, then relaunch or reload the nav nodes.
# The web "odom" button uses the first waypoint of this route as the fixed odom fallback start pose.
nav_route_task_file: map/routes/1hao_reall.json
nav_route_auto_align_enabled: false
nav_route_rotation_offset_deg: 0.0
nav_route_align_max_angle_deg: 6.0
nav_route_align_angle_step_deg: 0.5
nav_route_align_search_radius: 0.35
nav_avoid_regions_enabled: true
nav_avoid_region_margin: 0.0
# 0.0 means auto: use the nav_tools body+wheel lateral footprint for avoid-region inflation.
nav_avoid_footprint_radius: 0.0
nav_robot_body_length: 0.356
nav_robot_body_width: 0.235
nav_robot_body_center_x: 0.1518
nav_robot_origin_from_front: 0.105
nav_robot_pose_hip: 0.550
nav_robot_pose_knee: -1.125
nav_robot_wheel_vis_length: 0.16
nav_robot_wheel_vis_width: 0.055
nav_robot_footprint_padding: 0.02
odom_fallback_require_odom_fresh: true
odom_fallback_max_odom_age_ms: 500.0
odom_fallback_block_existing_map_odom_tf: true
odom_fallback_tf_conflict_window_s: 1.0
odom_fallback_tf_conflict_xy_tolerance: 0.05
odom_fallback_tf_conflict_yaw_tolerance_deg: 2.0
# Keep odom fallback running if Odin relocalizes mid-task; hand off after mission end or Exit odom.
odom_fallback_stop_on_external_tf: false
pcd_nav_file: map/1hao.pcd
pcd_floor_z_min: -1.6
pcd_floor_z_max: 0.4
pcd_sample_step: 25
pcd_robot_radius: 0.18
@@ -1,7 +1,7 @@
from launch import LaunchDescription
from launch.actions import DeclareLaunchArgument, IncludeLaunchDescription
from launch.launch_description_sources import PythonLaunchDescriptionSource
from launch.substitutions import LaunchConfiguration, PathJoinSubstitution, PythonExpression
from launch.substitutions import LaunchConfiguration, PathJoinSubstitution
from launch.conditions import IfCondition
from launch_ros.actions import Node
from launch_ros.parameter_descriptions import ParameterFile
@@ -27,7 +27,7 @@ def generate_launch_description():
launch_nav2_arg = DeclareLaunchArgument(
'launch_nav2',
default_value='false',
default_value='true',
description='Whether to launch the Nav2 navigation stack'
)
@@ -43,36 +43,6 @@ def generate_launch_description():
description='Whether to launch the Windows/Nano UDP web debug bridge'
)
launch_simple_nav_arg = DeclareLaunchArgument(
'launch_simple_nav',
default_value='true',
description='Whether to launch the simple waypoint navigation node'
)
localization_mode_arg = DeclareLaunchArgument(
'localization_mode',
default_value='relocal',
description='Localization profile: odom uses bridge fallback; relocal waits for Odin map/odom TF'
)
odin_config_file_arg = DeclareLaunchArgument(
'odin_config_file',
default_value=PathJoinSubstitution([
FindPackageShare('odin_ros_driver'),
'config',
'control_command_relocal.yaml',
]),
description='Odin control config YAML for the selected localization profile'
)
event_log_dir_arg = DeclareLaunchArgument(
'event_log_dir',
default_value=PythonExpression([
"'logs_v2_web/run_' + __import__('datetime').datetime.now().strftime('%Y-%m-%d_%H-%M-%S_%f')[:-3]"
]),
description='Per-run event log directory'
)
# Include odin_ros_driver launch
driver_launch = IncludeLaunchDescription(
PythonLaunchDescriptionSource(
@@ -82,10 +52,7 @@ def generate_launch_description():
'odin1_ros2.launch.py'
])
),
launch_arguments={
'launch_rviz': 'false',
'config_file': LaunchConfiguration('odin_config_file'),
}.items(),
launch_arguments={'launch_rviz': 'false'}.items(),
condition=IfCondition(LaunchConfiguration('launch_driver'))
)
@@ -106,23 +73,19 @@ def generate_launch_description():
launch_nav2_arg,
launch_remote_arg,
launch_web_bridge_arg,
launch_simple_nav_arg,
localization_mode_arg,
odin_config_file_arg,
event_log_dir_arg,
Node(
package="sim2real_hw",
executable="sim2real_hw_node",
name="sim2real_hw_node",
output="screen",
parameters=[runtime_params, {"event_log_dir": LaunchConfiguration("event_log_dir")}],
parameters=[runtime_params],
),
Node(
package="sim2real_runtime",
executable="sim2real_runtime_node",
name="sim2real_runtime_node",
output="screen",
parameters=[runtime_params, {"event_log_dir": LaunchConfiguration("event_log_dir")}],
parameters=[runtime_params],
),
Node(
package="sim2real_runtime",
@@ -136,7 +99,7 @@ def generate_launch_description():
executable="web_udp_bridge_node.py",
name="sim2real_web_udp_bridge_node",
output="screen",
parameters=[runtime_params, {"localization_mode": LaunchConfiguration("localization_mode")}],
parameters=[runtime_params],
condition=IfCondition(LaunchConfiguration('launch_web_bridge')),
),
Node(
@@ -147,14 +110,6 @@ def generate_launch_description():
parameters=[runtime_params],
condition=IfCondition(LaunchConfiguration('launch_remote')),
),
Node(
package="sim2real_runtime",
executable="simple_nav_node.py",
name="sim2real_simple_nav_node",
output="screen",
parameters=[runtime_params],
condition=IfCondition(LaunchConfiguration('launch_simple_nav')),
),
Node(
package="sim2real_runtime",
executable="odom_relay_node",
@@ -171,3 +126,4 @@ def generate_launch_description():
driver_launch,
nav2_launch,
])
@@ -19,8 +19,8 @@ struct DeploymentContract
static constexpr std::array<int, 4> kWheelIndices = {12, 13, 14, 15};
static constexpr float kLegKp = 50.0f;
static constexpr float kLegKd = 1.5f;
static constexpr float kLegHoldKp = kLegKp;
static constexpr float kLegHoldKd = kLegKd;
static constexpr float kLegHoldKp = 80.0f;
static constexpr float kLegHoldKd = 4.0f;
static constexpr float kWheelKd = 1.0f;
static constexpr std::array<int, 16> kCanBusMap = {
@@ -64,10 +64,10 @@ struct DeploymentContract
};
static constexpr std::array<float, 16> kDefaultDofPos = {
0.0f, 0.550f, -1.125f,
0.0f, 0.550f, -1.125f,
0.0f, 0.550f, -1.125f,
0.0f, 0.550f, -1.125f,
0.0f, 0.9f, -1.8f,
0.0f, 0.9f, -1.8f,
0.0f, 0.9f, -1.8f,
0.0f, 0.9f, -1.8f,
0.0f, 0.0f, 0.0f, 0.0f
};
};
@@ -1,11 +1,10 @@
#pragma once
#include <array>
#include <vector>
#include <cmath>
#include <algorithm>
#include "sim2real_common/deployment_contract.hpp"
#ifndef M_PI
#define M_PI 3.14159265358979323846
#endif
@@ -19,15 +18,12 @@ public:
StandBalanceController(double control_dt = 0.02)
: control_dt_(control_dt)
{
profile_h_ = {0.157f, 0.248f, 0.311f, 0.366f, 0.411f, 0.448f};
profile_hip_ = {1.5f, 1.2f, 1.0f, 0.8f, 0.6f, 0.4f};
profile_knee_ = {-2.5f, -2.1f, -1.8f, -1.5f, -1.2f, -0.9f};
reset();
}
void setNominalLegPose(float hip_pitch, float knee)
{
nominal_hip_pitch_ = hip_pitch;
nominal_knee_ = knee;
}
void reset()
{
stable_time_ = 0.0f;
@@ -38,8 +34,9 @@ public:
const std::array<float, 3>& imu_gyro,
const std::array<float, 3>& cmd)
{
const float hip_base = nominal_hip_pitch_;
const float knee_base = nominal_knee_;
float hip_base = 0.9f;
float knee_base = -1.8f;
estimateBaseLegPose(hip_base, knee_base);
float roll = 0.0f;
float pitch = 0.0f;
@@ -87,9 +84,28 @@ private:
pitch = std::atan2(gx, std::sqrt(std::max(1e-6f, gy * gy + gz * gz)));
}
void estimateBaseLegPose(float& hip, float& knee)
{
float h_clamp = std::clamp(height_, profile_h_.front(), profile_h_.back());
hip = interpolate(h_clamp, profile_h_, profile_hip_);
knee = interpolate(h_clamp, profile_h_, profile_knee_);
}
float interpolate(float x, const std::vector<float>& xp, const std::vector<float>& fp)
{
if (x <= xp.front()) return fp.front();
if (x >= xp.back()) return fp.back();
for (std::size_t i = 0; i < xp.size() - 1; ++i) {
if (x >= xp[i] && x <= xp[i+1]) {
float f = (x - xp[i]) / (xp[i+1] - xp[i]);
return fp[i] + f * (fp[i+1] - fp[i]);
}
}
return fp.back();
}
double control_dt_;
float nominal_hip_pitch_{DeploymentContract::kDefaultDofPos[1]};
float nominal_knee_{DeploymentContract::kDefaultDofPos[2]};
float height_{0.33f};
float kp_roll_{0.85f};
float kd_roll_rate_{0.03f};
float lateral_lean_gain_{0.0f};
@@ -101,6 +117,10 @@ private:
float stable_gyro_deg_s_{45.0f};
float enter_hold_s_{1.0f};
std::vector<float> profile_h_;
std::vector<float> profile_hip_;
std::vector<float> profile_knee_;
float stable_time_{0.0f};
};
@@ -12,41 +12,16 @@
#include "sensor_msgs/msg/imu.hpp"
#include "nav_msgs/msg/odometry.hpp"
#include "std_msgs/msg/bool.hpp"
#include "std_msgs/msg/string.hpp"
#include "sim2real_interfaces/msg/runtime_state.hpp"
#include "sim2real_interfaces/msg/runtime_target.hpp"
#include "sim2real_common/event_logger.hpp"
#include "sim2real_common/low_pass_filter.hpp"
#include "sim2real_common/mahony_filter.hpp"
#include "sim2real_common/safety_monitor.hpp"
#include "sim2real_common/runtime_guard.hpp"
struct can_frame;
namespace sim2real_hw
{
enum class RecoveryKind
{
None,
Stale,
NoEffect
};
enum class RecoveryStage
{
Idle,
AwaitInitFeedback,
AwaitEffectVerification
};
enum class ActiveModelMode
{
Rough,
Crawl,
Wall
};
struct MotorConfig
{
int bus; // 1 or 2
@@ -61,49 +36,14 @@ struct MotorStateInternal
float velocity{0.0f};
float torque{0.0f};
float temperature{0.0f};
float bus_voltage{0.0f};
float estimated_current_arms{0.0f};
float last_command_sim{0.0f};
std::uint32_t update_count{0};
std::uint32_t stale_count{0};
std::uint32_t command_active_count{0};
std::uint32_t no_effect_count{0};
std::uint16_t fault_code{0};
std::uint16_t fault_detail_1{0};
std::uint16_t fault_detail_2{0};
// Hold-over state
float last_valid_pos{0.0f};
float last_valid_vel{0.0f};
float last_valid_torque{0.0f};
std::uint32_t prev_update_count{0};
bool has_valid_data{false};
bool has_bus_voltage{false};
bool has_fault_snapshot{false};
bool stale_reported{false};
bool recovered_reported{false};
bool disable_reported{false};
bool command_effect_monitoring_active{false};
bool no_effect_reported{false};
bool high_temp_reported{false};
bool high_current_reported{false};
bool high_voltage_reported{false};
bool low_voltage_reported{false};
bool fault_code_reported{false};
bool init_confirmed{false};
std::uint32_t init_attempt_count{0};
std::uint32_t recovery_attempt_count{0};
std::uint32_t no_effect_recovery_attempt_count{0};
std::chrono::steady_clock::time_point last_recovery_attempt_time_{};
std::chrono::steady_clock::time_point last_no_effect_recovery_attempt_time_{};
std::chrono::steady_clock::time_point last_diag_snapshot_time_{};
std::chrono::steady_clock::time_point last_diag_request_time_{};
std::chrono::steady_clock::time_point recovery_stage_deadline_{};
std::uint32_t recovery_start_update_count{0};
std::uint32_t recovery_active_attempt_number{0};
RecoveryKind recovery_kind{RecoveryKind::None};
RecoveryStage recovery_stage{RecoveryStage::Idle};
std::string recovery_trigger;
std::string last_power_event_reason;
};
class HardwareBridgeNode : public rclcpp::Node
@@ -114,7 +54,6 @@ public:
private:
void onTarget(const sim2real_interfaces::msg::RuntimeTarget::SharedPtr msg);
void onModelStatus(const std_msgs::msg::String::SharedPtr msg);
void onReadLoop();
void onWriteLoop();
void onImu(const sensor_msgs::msg::Imu::SharedPtr msg);
@@ -125,83 +64,25 @@ private:
bool readCanFrame(int fd, void* frame, int timeout_us);
bool enableMotor(int fd, int motor_id);
bool disableMotor(int fd, int motor_id, bool clear_fault = false);
bool writeParameterInt(int fd, int motor_id, std::uint16_t param_id, std::uint32_t value);
bool disableMotor(int fd, int motor_id);
bool setModeRaw(int fd, int motor_id, std::int8_t mode);
bool readParameter(int fd, int motor_id, std::uint16_t param_id);
bool writeLimit(int fd, int motor_id, std::uint16_t param_id, float limit);
bool writeOperationFrame(int fd, int motor_id, double pos, double vel, double kp, double kd, double torque);
bool initializeMotor(std::size_t index, const std::string & reason, int max_attempts = 3);
bool initializeMotorsOnBus(int bus_id, const std::string & reason);
bool waitForMotorFeedback(std::size_t index, std::chrono::milliseconds timeout);
void processCanFrame(const struct can_frame & frame, int bus_id);
void drainCanFrames(int fd, int bus_id, int timeout_us);
bool isLegMotor(std::size_t index) const;
bool isWheelMotor(std::size_t index) const;
bool motorHasBlockingFault(std::size_t index) const;
bool isNoEffectConditionPresent(std::size_t index) const;
std::uint32_t noEffectCommandWarmupCycles(std::size_t index) const;
std::uint32_t noEffectTriggerCycles(std::size_t index) const;
std::uint32_t noEffectAttemptLimit(std::size_t index) const;
std::uint32_t noEffectCooldownMs(std::size_t index) const;
std::uint32_t noEffectVerifyTimeoutMs(std::size_t index) const;
bool hasFreshNoEffectDiagnostics(std::size_t index) const;
void requestMotorDiagnostics(std::size_t index);
std::string classifyNoEffectSuspect(std::size_t index) const;
std::string buildNoEffectSummary(std::size_t index) const;
void updateMotorCommandTracking(std::size_t index, float sim_command, const std::string & target_source);
void updateNoEffectDetection(std::size_t index);
bool startMotorRecoverySequence(
std::size_t index,
const std::string & trigger,
RecoveryKind kind,
std::uint32_t attempt_number);
void processMotorRecoverySequence(std::size_t index);
void clearMotorRecoverySequence(std::size_t index);
bool shouldAttemptMotorRecovery(std::size_t index) const;
bool attemptMotorRecovery(std::size_t index, const std::string & trigger);
bool shouldAttemptNoEffectRecovery(std::size_t index) const;
bool attemptNoEffectRecovery(std::size_t index, const std::string & trigger);
const char * jointName(std::size_t index) const;
std::string motorTag(std::size_t index) const;
float estimateCurrentArms(float torque_nm) const;
std::string decodeFaultCode(std::uint16_t fault_code) const;
std::string decodeFaultDetailRegister(std::uint16_t register_value, int register_index) const;
std::string buildMotorFaultSummary(std::size_t index) const;
std::string formatProtectionReason(const std::string & trigger, const std::string & reason) const;
void logProtectionEvent(const std::string & trigger, const std::string & reason, const std::string & action);
void logMotorPowerEvent(std::size_t index, const std::string & state, const std::string & reason);
void logMotorDiagnosticEvent(std::size_t index, const std::string & event, const std::string & reason, const char * level = "WARN");
void updateMotorTelemetry(std::size_t index, float pos_sim, float vel_sim, float torque_sim, float temperature_c);
void handleParameterResponse(const struct can_frame & frame, int bus_id);
void updateMotorDiagnostics(std::size_t index);
void pollMotorDiagnostics();
void finalizeRunSummary();
rclcpp::Publisher<sim2real_interfaces::msg::RuntimeState>::SharedPtr state_pub_;
rclcpp::Subscription<sim2real_interfaces::msg::RuntimeTarget>::SharedPtr target_sub_;
rclcpp::Subscription<std_msgs::msg::String>::SharedPtr model_status_sub_;
rclcpp::Subscription<sensor_msgs::msg::Imu>::SharedPtr imu_sub_;
rclcpp::Subscription<std_msgs::msg::Bool>::SharedPtr estop_sub_;
rclcpp::Subscription<nav_msgs::msg::Odometry>::SharedPtr odom_sub_;
rclcpp::TimerBase::SharedPtr read_timer_;
rclcpp::TimerBase::SharedPtr write_timer_;
rclcpp::CallbackGroup::SharedPtr motor_callback_group_;
rclcpp::CallbackGroup::SharedPtr sensor_callback_group_;
rclcpp::CallbackGroup::SharedPtr control_callback_group_;
std::mutex target_mutex_;
std::array<float, 16> latest_target_{};
std::array<float, 16> latest_raw_action_{};
std::string latest_target_source_{"boot_hold"};
rclcpp::Time latest_target_stamp_{0, 0, RCL_ROS_TIME};
std::array<float, 16> rough_default_dof_pos_{};
std::array<float, 16> crawl_default_dof_pos_{};
std::array<float, 16> wall_default_dof_pos_{};
std::array<float, 16> active_default_dof_pos_{};
ActiveModelMode active_model_mode_{ActiveModelMode::Rough};
bool model_switch_active_{false};
std::uint32_t target_sequence_{0};
std::uint32_t state_sequence_{0};
double target_timeout_ms_{150.0};
@@ -220,7 +101,6 @@ private:
// Hold-over constants
static constexpr std::uint32_t kHoldoverThreshold = 2;
static constexpr std::uint32_t kMotorDropReportThreshold = 40;
// Motor configurations and states
std::array<MotorConfig, 16> motors_;
@@ -253,71 +133,24 @@ private:
std::unique_ptr<sim2real_common::LowPassFilter> lpf_wheels_;
std::unique_ptr<sim2real_common::MahonyFilter> mahony_filter_;
std::unique_ptr<sim2real_common::SafetyMonitor> safety_monitor_;
std::unique_ptr<sim2real_common::SafetyMonitor> model_switch_safety_monitor_;
std::unique_ptr<sim2real_common::RuntimeGuard> runtime_guard_;
std::atomic<bool> mahony_initialized_{false};
bool mahony_initialized_{false};
rclcpp::Time last_read_time_{0, 0, RCL_ROS_TIME};
rclcpp::Time startup_soft_hold_start_time_{0, 0, RCL_ROS_TIME};
rclcpp::Time last_diag_poll_time_{0, 0, RCL_ROS_TIME};
std::size_t diag_poll_motor_index_{0};
// Telemetry
std::uint32_t fresh_count_{0};
std::uint32_t holdover_count_{0};
std::uint32_t stale_max_{0};
std::uint32_t holdover_events_total_{0};
std::uint32_t protection_trigger_count_{0};
std::uint32_t motor_drop_event_count_{0};
std::uint32_t motor_recover_event_count_{0};
std::uint32_t motor_fault_event_count_{0};
bool timeout_hold_logged_{false};
bool clip_active_logged_{false};
bool dry_run_{false};
std::atomic<bool> estop_triggered_{false};
std::atomic<bool> safety_enabled_{true};
std::atomic<bool> safety_triggered_{false};
std::string safety_reason_{""};
sim2real_common::EventLogger event_logger_;
std::string run_log_dir_;
float motor_temp_warn_c_{100.0f};
float motor_temp_fault_c_{135.0f};
float motor_bus_overvoltage_v_{60.0f};
float motor_bus_undervoltage_v_{12.0f};
float motor_current_warn_arms_{10.5f};
float motor_current_peak_arms_{14.0f};
float motor_torque_warn_nm_{13.0f};
float motor_torque_peak_nm_{17.0f};
double diag_poll_period_s_{0.10};
float wheel_no_effect_command_threshold_{1.0f};
float wheel_no_effect_min_response_ratio_{0.20f};
float wheel_no_effect_velocity_epsilon_{0.25f};
float wheel_no_effect_max_temperature_c_{90.0f};
float wheel_no_effect_min_bus_voltage_v_{18.0f};
std::uint32_t wheel_no_effect_command_warmup_cycles_{12};
std::uint32_t wheel_no_effect_trigger_cycles_{30};
std::uint32_t wheel_no_effect_attempt_limit_{2};
std::uint32_t wheel_no_effect_cooldown_ms_{1200};
std::uint32_t wheel_recovery_verify_timeout_ms_{180};
std::uint32_t wheel_no_effect_diag_freshness_ms_{350};
std::uint32_t wheel_no_effect_diag_request_period_ms_{80};
float leg_no_effect_position_error_threshold_{0.18f};
float leg_no_effect_velocity_epsilon_{0.12f};
float leg_no_effect_max_estimated_current_arms_{4.0f};
float leg_no_effect_max_abs_torque_nm_{5.0f};
float leg_no_effect_max_temperature_c_{100.0f};
float leg_no_effect_min_bus_voltage_v_{18.0f};
std::uint32_t leg_no_effect_command_warmup_cycles_{40};
std::uint32_t leg_no_effect_trigger_cycles_{25};
std::uint32_t leg_no_effect_attempt_limit_{2};
std::uint32_t leg_no_effect_cooldown_ms_{1200};
std::uint32_t leg_recovery_verify_timeout_ms_{220};
void onEstop(const std_msgs::msg::Bool::SharedPtr msg);
void logEvent(
const std::string & level,
const std::string & event,
const std::string & message);
};
} // namespace sim2real_hw
File diff suppressed because it is too large Load Diff
@@ -133,7 +133,7 @@ planner_server:
GridTransition:
plugin: "nav2_navfn_planner/NavfnPlanner"
tolerance: 0.5
use_astar: true
use_astar: false
allow_unknown: true
behavior_server:
@@ -2,7 +2,6 @@ cmake_minimum_required(VERSION 3.8)
project(sim2real_runtime)
find_package(ament_cmake REQUIRED)
find_package(CUDAToolkit QUIET)
find_package(geometry_msgs REQUIRED)
find_package(nav_msgs REQUIRED)
find_package(rclcpp REQUIRED)
@@ -32,33 +31,10 @@ find_library(ONNXRUNTIME_LIBRARY NAMES onnxruntime
get_filename_component(ONNXRUNTIME_LIBRARY_DIR ${ONNXRUNTIME_LIBRARY} DIRECTORY)
find_path(TENSORRT_INCLUDE_DIR NvInfer.h
PATHS
/usr/include
/usr/include/aarch64-linux-gnu
/usr/local/include
)
find_library(TENSORRT_LIBRARY NAMES nvinfer
PATHS
/usr/lib
/usr/lib/aarch64-linux-gnu
/usr/lib/x86_64-linux-gnu
/usr/local/lib
)
if(NOT ONNXRUNTIME_INCLUDE_DIR OR NOT ONNXRUNTIME_LIBRARY)
message(FATAL_ERROR "ONNX Runtime not found! Please install it or specify include/library paths.")
endif()
set(SIM2REAL_RUNTIME_HAS_TENSORRT FALSE)
if(TENSORRT_INCLUDE_DIR AND TENSORRT_LIBRARY AND CUDAToolkit_FOUND)
set(SIM2REAL_RUNTIME_HAS_TENSORRT TRUE)
message(STATUS "TensorRT support enabled for sim2real_runtime")
else()
message(STATUS "TensorRT support disabled for sim2real_runtime (missing TensorRT or CUDA toolkit)")
endif()
add_executable(sim2real_runtime_node
src/policy_runtime_node.cpp
)
@@ -67,60 +43,23 @@ add_executable(odom_relay_node
src/odom_relay_node.cpp
)
if(SIM2REAL_RUNTIME_HAS_TENSORRT)
add_executable(compare_onnx_trt
src/compare_onnx_trt.cpp
)
endif()
target_include_directories(sim2real_runtime_node PRIVATE
include
${ONNXRUNTIME_INCLUDE_DIR}
)
if(SIM2REAL_RUNTIME_HAS_TENSORRT)
target_include_directories(sim2real_runtime_node PRIVATE
${TENSORRT_INCLUDE_DIR}
)
target_include_directories(compare_onnx_trt PRIVATE
${TENSORRT_INCLUDE_DIR}
${ONNXRUNTIME_INCLUDE_DIR}
)
endif()
target_include_directories(odom_relay_node PRIVATE include)
target_link_libraries(sim2real_runtime_node
${ONNXRUNTIME_LIBRARY}
)
if(SIM2REAL_RUNTIME_HAS_TENSORRT)
target_link_libraries(sim2real_runtime_node
${TENSORRT_LIBRARY}
CUDA::cudart
)
target_link_libraries(compare_onnx_trt
${ONNXRUNTIME_LIBRARY}
${TENSORRT_LIBRARY}
CUDA::cudart
)
target_compile_definitions(sim2real_runtime_node PRIVATE
SIM2REAL_RUNTIME_HAS_TENSORRT=1
)
target_compile_definitions(compare_onnx_trt PRIVATE
SIM2REAL_RUNTIME_HAS_TENSORRT=1
)
endif()
set_target_properties(sim2real_runtime_node PROPERTIES
BUILD_RPATH "${ONNXRUNTIME_LIBRARY_DIR}"
INSTALL_RPATH "${ONNXRUNTIME_LIBRARY_DIR}"
)
target_compile_features(sim2real_runtime_node PRIVATE cxx_std_17)
if(SIM2REAL_RUNTIME_HAS_TENSORRT)
target_compile_features(compare_onnx_trt PRIVATE cxx_std_17)
endif()
ament_target_dependencies(sim2real_runtime_node
geometry_msgs
@@ -132,12 +71,6 @@ ament_target_dependencies(sim2real_runtime_node
sim2real_interfaces
)
if(SIM2REAL_RUNTIME_HAS_TENSORRT)
ament_target_dependencies(compare_onnx_trt
sim2real_common
)
endif()
ament_target_dependencies(odom_relay_node
geometry_msgs
nav_msgs
@@ -155,26 +88,11 @@ install(
DESTINATION lib/${PROJECT_NAME}
)
if(SIM2REAL_RUNTIME_HAS_TENSORRT)
install(
TARGETS compare_onnx_trt
DESTINATION lib/${PROJECT_NAME}
)
endif()
install(
PROGRAMS
src/remote_uart_node.py
src/cmd_mux_node.py
src/web_udp_bridge_node.py
src/simple_nav_node.py
src/pcd_nav_click_tool.py
DESTINATION lib/${PROJECT_NAME}
)
install(
FILES
src/deadzone_velocity_limiter.py
DESTINATION lib/${PROJECT_NAME}
)
@@ -5,18 +5,15 @@
#include <memory>
#include <string>
#include <vector>
#include <cstdint>
#include <atomic>
#include <chrono>
#include "geometry_msgs/msg/twist.hpp"
#include "geometry_msgs/msg/twist_stamped.hpp"
#include "std_msgs/msg/bool.hpp"
#include "std_msgs/msg/string.hpp"
#include "rclcpp/rclcpp.hpp"
#include "sim2real_interfaces/msg/runtime_state.hpp"
#include "sim2real_interfaces/msg/runtime_target.hpp"
#include "sim2real_common/event_logger.hpp"
#include "sim2real_common/stand_balance_controller.hpp"
#include "sim2real_common/safety_monitor.hpp"
#include "sim2real_common/runtime_guard.hpp"
@@ -24,11 +21,6 @@
// ONNXRuntime C++ API
#include <onnxruntime_cxx_api.h>
#ifdef SIM2REAL_RUNTIME_HAS_TENSORRT
#include <NvInfer.h>
#include <cuda_runtime_api.h>
#endif
namespace sim2real_runtime
{
@@ -36,86 +28,13 @@ class PolicyRuntimeNode : public rclcpp::Node
{
public:
PolicyRuntimeNode();
~PolicyRuntimeNode() override;
private:
enum class InferenceBackend {
None,
TensorRT,
OnnxRuntime,
};
enum class ModelMode {
Rough,
Crawl,
Wall,
};
enum class CrawlBackend {
Ik,
Rl,
};
enum class ModelSwitchState {
Idle,
ToStand,
StandHold,
ToModelPose,
};
enum class StartupState {
BOOT_HOLD,
STARTUP_SOFT_HOLD,
STARTUP_TRANSITION,
STARTUP_HOLD_AFTER,
RUNTIME
};
enum class PostureHoldMode {
None,
Keep,
ReturnDefault,
};
void onState(const sim2real_interfaces::msg::RuntimeState::SharedPtr msg);
void onCmdVel(const geometry_msgs::msg::Twist::SharedPtr msg);
void onCmdVelStamped(const geometry_msgs::msg::TwistStamped::SharedPtr msg);
void onModelSwitchCmd(const std_msgs::msg::String::SharedPtr msg);
void onPostureCmd(const std_msgs::msg::String::SharedPtr msg);
void applyCmdVel(float vx, float vy, float vyaw);
void onPolicyLoop();
bool initInferenceBackend();
bool initTensorRt();
bool initOnnxRuntime();
void shutdownOnnxRuntime();
void shutdownTensorRt();
std::string deriveTensorRtEnginePath(const std::string & onnx_model_path) const;
const std::array<float, 16> & defaultPoseForMode(ModelMode mode) const;
const std::string & modelPathForMode(ModelMode mode) const;
const std::string & modelEnginePathForMode(ModelMode mode) const;
bool switchInferenceModel(ModelMode target_mode);
void publishModelStatus();
bool modeUsesInference(ModelMode mode) const;
std::array<float, 16> computeHoldTarget(
const sim2real_interfaces::msg::RuntimeState & state,
const std::array<float, 3> & cmd);
std::array<float, 16> computeIkCrawlTarget(
const sim2real_interfaces::msg::RuntimeState & state,
const std::array<float, 3> & cmd);
float computeCrawlIkCommandScale(
const sim2real_interfaces::msg::RuntimeState & state,
const std::array<float, 16> & leg_target) const;
float projectedGravityTiltRad(const std::array<float, 3> & projected_gravity) const;
const char * startupStateName(StartupState state) const;
const char * modelModeName(ModelMode mode) const;
const char * modelSwitchStateName(ModelSwitchState state) const;
const char * postureHoldModeName(PostureHoldMode mode) const;
const char * inferenceBackendName() const;
const char * crawlBackendName() const;
void initializeDebugTrace();
void appendDebugTrace(
const sim2real_interfaces::msg::RuntimeState & state,
const sim2real_interfaces::msg::RuntimeTarget & target);
std::array<float, 53> buildObservation(
const sim2real_interfaces::msg::RuntimeState & state,
@@ -125,19 +44,11 @@ private:
std::array<float, 16> runPolicy(const std::array<float, 53> & obs);
bool isZeroCommand(const std::array<float, 3> & cmd, const std::array<float, 3> & imu_gyro) const;
bool isCommandActive(const std::array<float, 3> & cmd) const;
void startPostureTransition(
PostureHoldMode mode,
const std::array<float, 16> & start_pose,
const std::array<float, 16> & target_pose,
const rclcpp::Time & now_time);
rclcpp::Publisher<sim2real_interfaces::msg::RuntimeTarget>::SharedPtr target_pub_;
rclcpp::Publisher<std_msgs::msg::String>::SharedPtr model_status_pub_;
rclcpp::Subscription<sim2real_interfaces::msg::RuntimeState>::SharedPtr state_sub_;
rclcpp::Subscription<geometry_msgs::msg::Twist>::SharedPtr cmd_sub_;
rclcpp::Subscription<geometry_msgs::msg::TwistStamped>::SharedPtr cmd_stamped_sub_;
rclcpp::Subscription<std_msgs::msg::String>::SharedPtr model_switch_sub_;
rclcpp::Subscription<std_msgs::msg::String>::SharedPtr posture_cmd_sub_;
rclcpp::TimerBase::SharedPtr policy_timer_;
std::mutex mutex_;
@@ -150,6 +61,13 @@ private:
std::uint32_t sequence_{0};
// Startup State Machine
enum class StartupState {
BOOT_HOLD,
STARTUP_SOFT_HOLD,
STARTUP_TRANSITION,
STARTUP_HOLD_AFTER,
RUNTIME
};
StartupState startup_state_{StartupState::BOOT_HOLD};
std::array<float, 16> start_pose_{};
std::array<float, 16> startup_delta_{};
@@ -158,65 +76,9 @@ private:
double hold_time_{1.0};
std::unique_ptr<sim2real_common::StandBalanceController> stand_balance_;
ModelMode current_model_mode_{ModelMode::Rough};
ModelMode requested_model_mode_{ModelMode::Rough};
ModelMode loaded_model_mode_{ModelMode::Rough};
CrawlBackend crawl_backend_{CrawlBackend::Ik};
ModelSwitchState model_switch_state_{ModelSwitchState::Idle};
bool model_switch_requested_{false};
bool hold_active_model_pose_when_unreleased_{false};
std::array<float, 16> rough_default_dof_pos_{};
std::array<float, 16> crawl_default_dof_pos_{};
std::array<float, 16> wall_default_dof_pos_{};
std::array<float, 16> active_default_dof_pos_{};
std::array<float, 16> safety_reference_dof_pos_{};
std::array<float, 16> keep_pose_dof_pos_{};
std::array<float, 16> posture_start_pose_{};
std::array<float, 16> posture_target_pose_{};
std::array<float, 16> posture_delta_{};
PostureHoldMode posture_hold_mode_{PostureHoldMode::None};
bool posture_transition_active_{false};
rclcpp::Time posture_transition_start_time_{0, 0, RCL_ROS_TIME};
double posture_transition_s_{0.8};
std::array<float, 16> switch_start_pose_{};
std::array<float, 16> switch_delta_{};
rclcpp::Time model_switch_state_start_time_{0, 0, RCL_ROS_TIME};
double model_switch_transition_s_{1.2};
double model_switch_to_stand_transition_scale_{1.35};
double model_switch_to_model_transition_scale_{1.55};
double model_switch_min_transition_s_{0.35};
double model_switch_stand_hold_s_{0.45};
double model_switch_stand_max_err_{0.18};
double model_switch_stand_max_vel_{0.8};
double active_switch_transition_s_{1.2};
std::string rough_model_path_{"policies/model_rough.onnx"};
std::string rough_model_engine_path_{""};
std::string crawl_model_path_{"policies/model_crawl.onnx"};
std::string crawl_model_engine_path_{""};
std::string wall_model_path_{"policies/model_wall.onnx"};
std::string wall_model_engine_path_{""};
float crawl_ik_wheel_linear_gain_{6.25f};
float crawl_ik_wheel_yaw_gain_{4.0f};
float crawl_ik_max_wheel_speed_{6.0f};
float crawl_ik_abduction_clip_{0.45f};
float crawl_ik_yaw_rate_kp_{0.0f};
bool crawl_ik_imu_posture_{false};
float crawl_ik_encoder_posture_kp_{0.0f};
float crawl_ik_encoder_posture_max_{0.03f};
bool crawl_ik_encoder_guard_{true};
float crawl_ik_encoder_guard_start_{0.28f};
float crawl_ik_encoder_guard_stop_{0.65f};
bool crawl_ik_imu_guard_{true};
float crawl_ik_imu_guard_start_rad_{0.20943952f};
float crawl_ik_imu_guard_stop_rad_{0.48869219f};
// ONNX Runtime members
std::string model_path_{"policies/model_rough.onnx"};
std::string model_engine_path_{""};
bool prefer_tensorrt_{true};
bool use_cuda_{false}; // enable CUDA Execution Provider on Orin Nano
InferenceBackend inference_backend_{InferenceBackend::None};
std::unique_ptr<Ort::Env> env_;
std::unique_ptr<Ort::Session> session_;
std::unique_ptr<Ort::MemoryInfo> memory_info_;
@@ -229,35 +91,19 @@ private:
std::vector<std::int64_t> input_shape_;
std::vector<std::int64_t> output_shape_;
#ifdef SIM2REAL_RUNTIME_HAS_TENSORRT
nvinfer1::IRuntime * trt_runtime_{nullptr};
nvinfer1::ICudaEngine * trt_engine_{nullptr};
nvinfer1::IExecutionContext * trt_context_{nullptr};
cudaStream_t trt_stream_{nullptr};
void * trt_input_buffer_{nullptr};
void * trt_output_buffer_{nullptr};
std::string trt_input_name_;
std::string trt_output_name_;
#endif
// Command filter and release states
std::array<float, 3> filtered_cmd_{{0.0f, 0.0f, 0.0f}};
float runtime_max_vx_{1.0f};
float runtime_max_vy_{0.3f};
float runtime_max_yaw_rate_{1.0f};
float release_alpha_{0.0f};
float command_release_s_{0.35f};
float release_command_hold_s_{0.12f};
float release_posture_max_err_{0.35f};
float release_target_blend_s_{0.30f};
float model_switch_release_scale_{1.3f};
float clip_obs_{100.0f};
bool hold_zero_command_pose_{true};
bool enable_zero_cmd_suppression_{true};
bool require_active_command_to_release_{true};
bool zero_cmd_use_yaw_rate_{false};
bool runtime_released_{false};
bool slow_release_after_model_switch_{false};
float release_active_time_{0.0f};
float zero_cmd_lin_thresh_{0.05f};
float zero_cmd_yaw_thresh_{0.05f};
@@ -269,29 +115,10 @@ private:
std::atomic<bool> safety_enabled_{true};
std::atomic<bool> safety_triggered_{false};
std::string safety_reason_{""};
sim2real_common::EventLogger event_logger_;
std::string run_log_dir_;
std::string debug_trace_path_;
bool debug_trace_enabled_{true};
std::uint32_t debug_trace_decimation_{1};
std::uint32_t debug_trace_counter_{0};
std::uint32_t protection_trigger_count_{0};
std::uint32_t target_clip_count_{0};
bool clip_active_logged_{false};
std::unique_ptr<sim2real_common::SafetyMonitor> safety_monitor_;
std::unique_ptr<sim2real_common::SafetyMonitor> model_switch_safety_monitor_;
std::unique_ptr<sim2real_common::RuntimeGuard> runtime_guard_;
void onEstop(const std_msgs::msg::Bool::SharedPtr msg);
void logEvent(
const std::string & level,
const std::string & event,
const std::string & message);
void logProtectionEvent(
const std::string & trigger,
const std::string & reason,
const std::string & action);
void finalizeRunSummary();
};
} // namespace sim2real_runtime
@@ -12,13 +12,10 @@
<depend>nav_msgs</depend>
<depend>rclcpp</depend>
<depend>std_msgs</depend>
<depend>tf2_msgs</depend>
<depend>tf2_ros</depend>
<depend>sim2real_common</depend>
<depend>sim2real_interfaces</depend>
<exec_depend>python3-matplotlib</exec_depend>
<exec_depend>python3-serial</exec_depend>
<exec_depend>python3-yaml</exec_depend>
<exec_depend>rclpy</exec_depend>
<export>
@@ -10,15 +10,12 @@ from rclpy.executors import ExternalShutdownException
from rclpy.node import Node
from std_msgs.msg import Bool, String
from deadzone_velocity_limiter import limit_deadzone_axis
class ControlMode(str, Enum):
DISABLED = "DISABLED"
REMOTE = "REMOTE"
WEB = "WEB"
NAV = "NAV"
KEEP = "KEEP"
class CmdMuxNode(Node):
@@ -36,37 +33,6 @@ class CmdMuxNode(Node):
self.max_vx_acc = float(self.declare_parameter("cmd_mux_max_vx_acc", 1.0).value)
self.max_vy_acc = float(self.declare_parameter("cmd_mux_max_vy_acc", 1.0).value)
self.max_yaw_acc = float(self.declare_parameter("cmd_mux_max_yaw_acc", 1.5).value)
self.max_vx_decel = float(
self.declare_parameter("cmd_mux_max_vx_decel", self.max_vx_acc).value
)
self.max_vy_decel = float(
self.declare_parameter("cmd_mux_max_vy_decel", self.max_vy_acc).value
)
self.max_yaw_decel = float(
self.declare_parameter("cmd_mux_max_yaw_decel", self.max_yaw_acc).value
)
self.linear_deadzone_epsilon = float(
self.declare_parameter("cmd_mux_linear_deadzone_epsilon", 0.0).value
)
self.yaw_deadzone_epsilon = float(
self.declare_parameter("cmd_mux_yaw_deadzone_epsilon", 0.0).value
)
self.min_effective_vx = float(
self.declare_parameter("cmd_mux_min_effective_vx", 0.0).value
)
self.min_effective_vy = float(
self.declare_parameter("cmd_mux_min_effective_vy", 0.0).value
)
self.min_effective_yaw = float(
self.declare_parameter("cmd_mux_min_effective_yaw_rate", 0.0).value
)
self.deadzone_sources = {
item.strip().lower()
for item in str(
self.declare_parameter("cmd_mux_deadzone_sources", "nav").value
).split(",")
if item.strip()
}
self.mode = self.parse_mode(self.default_mode)
self.estop = False
@@ -163,16 +129,9 @@ class CmdMuxNode(Node):
elif self.mode == ControlMode.NAV and self.nav_enabled and self.is_fresh(self.nav_stamp, self.nav_timeout_ms, now):
target = self.latest_nav
source = "nav"
elif self.mode == ControlMode.KEEP:
source = "keep"
target = self.limit_twist(target)
if self.estop:
target = Twist()
self.last_output = Twist()
self.last_pub_time = now
else:
target = self.accel_limit(target, now, source in self.deadzone_sources)
target = self.accel_limit(target, now)
self.cmd_pub.publish(target)
self.mode_pub.publish(String(data=self.mode.value))
self.status_pub.publish(String(data=f"mode={self.mode.value},source={source},estop={self.estop}"))
@@ -190,46 +149,12 @@ class CmdMuxNode(Node):
out.angular.z = self.clamp(msg.angular.z, -self.max_yaw, self.max_yaw)
return out
def accel_limit(
self,
target: Twist,
now: rclpy.time.Time,
apply_deadzone: bool,
) -> Twist:
def accel_limit(self, target: Twist, now: rclpy.time.Time) -> Twist:
dt = max((now - self.last_pub_time).nanoseconds / 1.0e9, 1.0e-3)
min_effective_vx = self.min_effective_vx if apply_deadzone else 0.0
min_effective_vy = self.min_effective_vy if apply_deadzone else 0.0
min_effective_yaw = self.min_effective_yaw if apply_deadzone else 0.0
linear_deadzone_epsilon = self.linear_deadzone_epsilon if apply_deadzone else 0.0
yaw_deadzone_epsilon = self.yaw_deadzone_epsilon if apply_deadzone else 0.0
out = Twist()
out.linear.x = limit_deadzone_axis(
self.last_output.linear.x,
target.linear.x,
dt,
self.max_vx_acc,
self.max_vx_decel,
min_effective_vx,
linear_deadzone_epsilon,
)
out.linear.y = limit_deadzone_axis(
self.last_output.linear.y,
target.linear.y,
dt,
self.max_vy_acc,
self.max_vy_decel,
min_effective_vy,
linear_deadzone_epsilon,
)
out.angular.z = limit_deadzone_axis(
self.last_output.angular.z,
target.angular.z,
dt,
self.max_yaw_acc,
self.max_yaw_decel,
min_effective_yaw,
yaw_deadzone_epsilon,
)
out.linear.x = self.step(self.last_output.linear.x, target.linear.x, self.max_vx_acc * dt)
out.linear.y = self.step(self.last_output.linear.y, target.linear.y, self.max_vy_acc * dt)
out.angular.z = self.step(self.last_output.angular.z, target.angular.z, self.max_yaw_acc * dt)
self.last_output = out
self.last_pub_time = now
return out
@@ -238,6 +163,16 @@ class CmdMuxNode(Node):
def clamp(value: float, low: float, high: float) -> float:
return max(low, min(high, float(value)))
@staticmethod
def step(current: float, target: float, max_delta: float) -> float:
delta = target - current
if delta > max_delta:
return current + max_delta
if delta < -max_delta:
return current - max_delta
return target
def main(args: Optional[list[str]] = None) -> None:
rclpy.init(args=args)
node = CmdMuxNode()
@@ -1,64 +0,0 @@
#!/usr/bin/env python3
from __future__ import annotations
import math
def step_towards(current: float, target: float, max_delta: float) -> float:
delta = target - current
if delta > max_delta:
return current + max_delta
if delta < -max_delta:
return current - max_delta
return target
def shape_deadzone_target(target: float, min_effective: float, zero_epsilon: float) -> float:
if abs(target) <= zero_epsilon:
return 0.0
if min_effective > 0.0 and abs(target) < min_effective:
return math.copysign(min_effective, target)
return target
def limit_deadzone_axis(
current: float,
target: float,
dt: float,
max_accel: float,
max_decel: float,
min_effective: float,
zero_epsilon: float,
) -> float:
dt = max(float(dt), 1.0e-3)
max_accel = max(float(max_accel), 0.0)
max_decel = max(float(max_decel), 0.0)
min_effective = max(float(min_effective), 0.0)
zero_epsilon = max(float(zero_epsilon), 0.0)
shaped_target = shape_deadzone_target(target, min_effective, zero_epsilon)
if min_effective <= 0.0:
accelerating = current * shaped_target >= 0.0 and abs(shaped_target) > abs(current)
limit = max_accel if accelerating else max_decel
return step_towards(current, shaped_target, limit * dt)
effective_current = 0.0 if abs(current) < min_effective else current
if shaped_target == 0.0:
stepped = step_towards(effective_current, 0.0, max_decel * dt)
return 0.0 if abs(stepped) < min_effective else stepped
if effective_current == 0.0:
return math.copysign(min_effective, shaped_target)
if effective_current * shaped_target < 0.0:
stepped = step_towards(effective_current, 0.0, max_decel * dt)
if stepped == 0.0 or abs(stepped) < min_effective:
return 0.0
return stepped
accelerating = abs(shaped_target) > abs(effective_current)
limit = max_accel if accelerating else max_decel
stepped = step_towards(effective_current, shaped_target, limit * dt)
if abs(stepped) < min_effective:
return math.copysign(min_effective, shaped_target)
return stepped
File diff suppressed because it is too large Load Diff
@@ -10,7 +10,7 @@ import rclpy
from geometry_msgs.msg import Twist
from rclpy.executors import ExternalShutdownException
from rclpy.node import Node
from std_msgs.msg import Bool, String
from std_msgs.msg import Bool
SBUS_FRAME_SIZE = 25
SBUS_RC_MID = 1024
@@ -23,22 +23,7 @@ SWITCH_HIGH = 1
@dataclass
class RemoteSwitchState:
ch5: int = SWITCH_MID
ch6: int = SWITCH_MID
ch7: int = SWITCH_MID
ch8: int = SWITCH_MID
ch9: int = SWITCH_MID
ch10: int = SWITCH_MID
def get(self, channel: int) -> Optional[int]:
return {
5: self.ch5,
6: self.ch6,
7: self.ch7,
8: self.ch8,
9: self.ch9,
10: self.ch10,
}.get(int(channel))
@dataclass
@@ -129,14 +114,7 @@ class SbusUartReceiver:
ch2=self._normalize_axis(channels[1]),
ch3=self._normalize_axis(channels[3]),
ch4=self._normalize_axis(channels[2]),
switches=RemoteSwitchState(
ch5=self._decode_switch(channels[4]),
ch6=self._decode_switch(channels[5]),
ch7=self._decode_switch(channels[6]),
ch8=self._decode_switch(channels[7]),
ch9=self._decode_switch(channels[8]),
ch10=self._decode_switch(channels[9]),
),
switches=RemoteSwitchState(ch7=self._decode_switch(channels[6])),
frame_ok=True,
)
if any(abs(value) > 800 for value in (state.ch1, state.ch2, state.ch3, state.ch4)):
@@ -176,31 +154,12 @@ class RemoteUartNode(Node):
self.publish_inactive_zero = bool(self.declare_parameter("remote_publish_inactive_zero", True).value)
self.estop_latch = bool(self.declare_parameter("remote_estop_latch", True).value)
self.poll_hz = float(self.declare_parameter("remote_poll_hz", 50.0).value)
self.default_mode = str(self.declare_parameter("cmd_mux_default_mode", "REMOTE").value).strip().upper()
self.model_switch_enabled = bool(self.declare_parameter("remote_model_switch_enabled", True).value)
self.model_switch_channel = int(self.declare_parameter("remote_model_switch_channel", 10).value)
self.model_switch_debounce_frames = max(int(self.declare_parameter("remote_model_switch_debounce_frames", 3).value), 1)
self.model_switch_rough_level = self.parse_switch_level(
str(self.declare_parameter("remote_model_switch_rough_level", "low").value)
)
legacy_ik_level = str(self.declare_parameter("remote_model_switch_crawl_level", "").value).strip()
ik_level_default = legacy_ik_level if legacy_ik_level else "high"
self.model_switch_ik_level = self.parse_switch_level(
str(self.declare_parameter("remote_model_switch_ik_level", ik_level_default).value)
)
self.cmd_pub = self.create_publisher(Twist, "cmd_vel_remote", 10)
self.estop_pub = self.create_publisher(Bool, "/safety/estop", 10)
self.model_cmd_pub = self.create_publisher(String, "runtime/model_cmd", 10)
self.receiver: Optional[SbusUartReceiver] = None
self.estop_published = False
self.open_error_logged = False
self.remote_mode_active = self.default_mode == "REMOTE"
self.model_switch_candidate: Optional[int] = None
self.model_switch_candidate_count = 0
self.model_switch_stable: Optional[int] = None
self.create_subscription(String, "control/mode_state", self.on_mode_state, 10)
if self.enabled:
self.receiver = SbusUartReceiver(
@@ -254,8 +213,6 @@ class RemoteUartNode(Node):
self.estop_pub.publish(Bool(data=False))
self.estop_published = False
self.handle_model_switch(state)
active = any(abs(value) > self.active_threshold for value in (state.ch1, state.ch2, state.ch4))
if active or self.publish_inactive_zero:
cmd = Twist()
@@ -267,15 +224,6 @@ class RemoteUartNode(Node):
def publish_zero_cmd(self) -> None:
self.cmd_pub.publish(Twist())
def on_mode_state(self, msg: String) -> None:
mode = str(msg.data).strip().upper()
remote_mode_active = mode == "REMOTE"
if remote_mode_active == self.remote_mode_active:
return
self.remote_mode_active = remote_mode_active
self.reset_model_switch_tracking()
def axis_to_velocity(self, raw_value: int, limit: float, invert: bool) -> float:
if abs(raw_value) <= self.active_threshold:
return 0.0
@@ -284,53 +232,6 @@ class RemoteUartNode(Node):
scaled = -scaled
return float(scaled * limit)
@staticmethod
def parse_switch_level(value: str) -> int:
normalized = value.strip().lower()
if normalized == "low":
return SWITCH_LOW
if normalized == "high":
return SWITCH_HIGH
return SWITCH_MID
def handle_model_switch(self, state: RemoteControlState) -> None:
if not self.model_switch_enabled or not self.remote_mode_active:
return
switch_level = state.switches.get(self.model_switch_channel)
if switch_level is None:
return
if switch_level == self.model_switch_candidate:
self.model_switch_candidate_count += 1
else:
self.model_switch_candidate = switch_level
self.model_switch_candidate_count = 1
if self.model_switch_candidate_count < self.model_switch_debounce_frames:
return
if switch_level == self.model_switch_stable:
return
self.model_switch_stable = switch_level
if switch_level == self.model_switch_rough_level:
self.model_cmd_pub.publish(String(data="rough"))
self.get_logger().info(
f"Remote model switch: CH{self.model_switch_channel} -> rough"
)
elif switch_level == self.model_switch_ik_level:
self.model_cmd_pub.publish(String(data="ik"))
self.get_logger().info(
f"Remote model switch: CH{self.model_switch_channel} -> ik"
)
def reset_model_switch_tracking(self) -> None:
self.model_switch_candidate = None
self.model_switch_candidate_count = 0
self.model_switch_stable = None
def main(args: Optional[list[str]] = None) -> None:
rclpy.init(args=args)
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
@@ -1,48 +0,0 @@
#!/usr/bin/env python3
from __future__ import annotations
import math
import sys
import unittest
from pathlib import Path
SOURCE_DIR = Path(__file__).resolve().parents[1] / "src"
sys.path.insert(0, str(SOURCE_DIR))
from deadzone_velocity_limiter import limit_deadzone_axis
class DeadzoneVelocityLimiterTest(unittest.TestCase):
def test_start_skips_linear_deadzone(self) -> None:
output = limit_deadzone_axis(0.0, 0.5, 0.02, 1.0, 2.0, 0.22, 0.05)
self.assertAlmostEqual(output, 0.22)
def test_small_nonzero_target_is_promoted(self) -> None:
output = limit_deadzone_axis(0.0, 0.1, 0.02, 1.0, 2.0, 0.22, 0.05)
self.assertAlmostEqual(output, 0.22)
def test_tiny_target_remains_zero(self) -> None:
output = limit_deadzone_axis(0.0, 0.04, 0.02, 1.0, 2.0, 0.22, 0.05)
self.assertEqual(output, 0.0)
def test_target_at_epsilon_remains_zero(self) -> None:
output = limit_deadzone_axis(0.0, 0.05, 0.02, 1.0, 2.0, 0.22, 0.05)
self.assertEqual(output, 0.0)
def test_stop_does_not_hold_minimum_speed(self) -> None:
output = limit_deadzone_axis(0.22, 0.0, 0.02, 1.0, 2.0, 0.22, 0.05)
self.assertEqual(output, 0.0)
def test_reverse_brakes_through_zero_first(self) -> None:
output = limit_deadzone_axis(0.30, -0.5, 0.02, 1.0, 2.0, 0.22, 0.05)
self.assertGreaterEqual(output, 0.0)
self.assertLess(output, 0.30)
def test_deadzone_disabled_preserves_legacy_ramp(self) -> None:
output = limit_deadzone_axis(0.0, 0.5, 0.02, 1.0, 1.0, 0.0, 0.0)
self.assertTrue(math.isclose(output, 0.02))
if __name__ == "__main__":
unittest.main()
@@ -0,0 +1,107 @@
#!/usr/bin/env python3
from __future__ import annotations
import argparse
import json
import socket
import threading
import time
from http.server import SimpleHTTPRequestHandler, ThreadingHTTPServer
from pathlib import Path
from typing import Optional
STATE_LOCK = threading.Lock()
LATEST_STATE: dict = {"type": "state", "connected": False}
NANO_ADDR: tuple[str, int]
UDP_SOCK: socket.socket
class Handler(SimpleHTTPRequestHandler):
def do_GET(self) -> None:
if self.path == "/api/state":
with STATE_LOCK:
data = json.dumps(LATEST_STATE).encode("utf-8")
self._json(200, data)
return
super().do_GET()
def do_POST(self) -> None:
if self.path not in ("/api/control", "/api/state"):
self.send_error(404)
return
length = int(self.headers.get("Content-Length", "0"))
body = self.rfile.read(length) if length else b"{}"
try:
payload = json.loads(body.decode("utf-8"))
send_udp(payload)
self._json(200, b'{"ok":true}')
except Exception as exc:
self._json(400, json.dumps({"ok": False, "error": str(exc)}).encode())
def _json(self, code: int, data: bytes) -> None:
self.send_response(code)
self.send_header("Content-Type", "application/json")
self.send_header("Content-Length", str(len(data)))
self.send_header("Access-Control-Allow-Origin", "*")
self.end_headers()
self.wfile.write(data)
def log_message(self, format: str, *args: object) -> None:
return
def send_udp(payload: dict) -> None:
data = json.dumps(payload, separators=(",", ":")).encode("utf-8")
UDP_SOCK.sendto(data, NANO_ADDR)
def udp_rx_loop(sock: socket.socket) -> None:
global LATEST_STATE
while True:
try:
data, _ = sock.recvfrom(65535)
payload = json.loads(data.decode("utf-8"))
payload["connected"] = True
payload["local_receive_time"] = time.time()
with STATE_LOCK:
LATEST_STATE = payload
except Exception:
time.sleep(0.01)
def heartbeat_loop() -> None:
while True:
try:
send_udp({"type": "ping", "stamp": time.time()})
except Exception:
pass
time.sleep(0.5)
def main() -> None:
global NANO_ADDR, UDP_SOCK
parser = argparse.ArgumentParser(description="Windows local web debug UI for sim2real_ros2")
parser.add_argument("--nano-host", required=True, help="Nano IP address")
parser.add_argument("--nano-port", type=int, default=15000, help="Nano UDP listen port")
parser.add_argument("--listen-host", default="0.0.0.0", help="Local HTTP host")
parser.add_argument("--http-port", type=int, default=8088, help="Local HTTP port")
parser.add_argument("--udp-port", type=int, default=15001, help="Local UDP receive port")
args = parser.parse_args()
NANO_ADDR = (args.nano_host, args.nano_port)
UDP_SOCK = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
UDP_SOCK.bind(("0.0.0.0", args.udp_port))
threading.Thread(target=udp_rx_loop, args=(UDP_SOCK,), daemon=True).start()
threading.Thread(target=heartbeat_loop, daemon=True).start()
static_dir = Path(__file__).resolve().parent / "static"
handler = lambda *a, **kw: Handler(*a, directory=str(static_dir), **kw)
httpd = ThreadingHTTPServer((args.listen_host, args.http_port), handler)
print(f"Open http://127.0.0.1:{args.http_port}")
print(f"UDP Nano={args.nano_host}:{args.nano_port} local={args.udp_port}")
httpd.serve_forever()
if __name__ == "__main__":
main()
@@ -0,0 +1,259 @@
'use strict';
const JOINT_NAMES = [
'FL_H_ABD','FL_H_PIT','FL_KNEE',
'FR_H_ABD','FR_H_PIT','FR_KNEE',
'RL_H_ABD','RL_H_PIT','RL_KNEE',
'RR_H_ABD','RR_H_PIT','RR_KNEE',
'FL_WHEEL','FR_WHEEL','RL_WHEEL','RR_WHEEL',
];
const $ = id => document.getElementById(id);
const cmd = { vx: 0, vy: 0, yaw: 0 };
let cmdSendTimer = null;
let currentMode = 'UNKNOWN';
// ── API ──────────────────────────────────────────────────────────────────────
async function post(payload) {
try {
await fetch('/api/control', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify(payload),
});
} catch (e) {
appendEvent('API_ERROR', e.message, 'bad');
}
}
function sendCmd() {
post({
type: 'cmd_vel',
linear: { x: cmd.vx, y: cmd.vy, z: 0 },
angular: { x: 0, y: 0, z: cmd.yaw },
});
$('cmd-display').textContent =
`vx=${cmd.vx.toFixed(2)} vy=${cmd.vy.toFixed(2)} yaw=${cmd.yaw.toFixed(2)}`;
}
function zeroAll() {
cmd.vx = 0; cmd.vy = 0; cmd.yaw = 0;
$('cmd-vx').value = 0;
$('cmd-vy').value = 0;
$('cmd-yaw').value = 0;
$('cmd-vx-v').textContent = '0.00';
$('cmd-vy-v').textContent = '0.00';
$('cmd-yaw-v').textContent = '0.00';
$('cmd-display').textContent = 'vx=0.00 vy=0.00 yaw=0.00';
$('stick').style.transform = 'translate(-50%, -50%)';
post({ type: 'zero' });
}
// ── Buttons ──────────────────────────────────────────────────────────────────
function setMode(mode) {
if (mode === 'WEB' && !confirm('确认切换到 WEB 控制?\n请确认机器人安全且速度为 0。')) return;
post({ type: 'mode', mode });
appendEvent('MODE_SET', `${mode}`, 'ok');
}
function highlightMode(mode) {
for (const m of ['DISABLED', 'REMOTE', 'WEB', 'NAV']) {
const btn = $('btn-' + m.toLowerCase());
if (btn) btn.classList.toggle('active-mode', m === mode);
}
const el = $('stage');
if (el) {
el.textContent = mode;
el.className = 'stage ' + mode;
}
currentMode = mode;
}
$('btn-disabled').onclick = () => { zeroAll(); setMode('DISABLED'); };
$('btn-remote').onclick = () => setMode('REMOTE');
$('btn-web').onclick = () => setMode('WEB');
$('btn-nav').onclick = () => setMode('NAV');
$('btn-zero').onclick = zeroAll;
$('btn-estop').onclick = () => {
if (confirm('确认触发软急停?')) {
post({ type: 'estop', data: true });
zeroAll();
appendEvent('ESTOP', '软急停已触发', 'bad');
}
};
// ── Sliders ──────────────────────────────────────────────────────────────────
for (const [id, key] of [['cmd-vx','vx'],['cmd-vy','vy'],['cmd-yaw','yaw']]) {
$(id).addEventListener('input', e => {
cmd[key] = parseFloat(e.target.value);
$(id + '-v').textContent = cmd[key].toFixed(2);
if (currentMode === 'WEB') sendCmd();
});
}
// ── Joystick ─────────────────────────────────────────────────────────────────
const joystick = $('joystick');
const stick = $('stick');
let dragging = false;
function updateJoystick(clientX, clientY) {
const rect = joystick.getBoundingClientRect();
const cx = rect.left + rect.width / 2;
const cy = rect.top + rect.height / 2;
const maxR = rect.width * 0.42;
let dx = clientX - cx;
let dy = clientY - cy;
const dist = Math.hypot(dx, dy);
if (dist > maxR) { dx = dx / dist * maxR; dy = dy / dist * maxR; }
stick.style.transform = `translate(calc(-50% + ${dx}px), calc(-50% + ${dy}px))`;
cmd.vx = parseFloat((-(dy / maxR) * 0.8).toFixed(3));
cmd.vy = parseFloat(( (dx / maxR) * 0.3).toFixed(3));
$('cmd-vx').value = cmd.vx;
$('cmd-vy').value = cmd.vy;
$('cmd-vx-v').textContent = cmd.vx.toFixed(2);
$('cmd-vy-v').textContent = cmd.vy.toFixed(2);
if (currentMode === 'WEB') sendCmd();
}
joystick.addEventListener('pointerdown', e => {
dragging = true;
joystick.setPointerCapture(e.pointerId);
updateJoystick(e.clientX, e.clientY);
});
joystick.addEventListener('pointermove', e => { if (dragging) updateJoystick(e.clientX, e.clientY); });
joystick.addEventListener('pointerup', () => { dragging = false; zeroAll(); });
joystick.addEventListener('pointercancel', () => { dragging = false; zeroAll(); });
// ── Joints grid init ─────────────────────────────────────────────────────────
function initJointsGrid() {
const grid = $('joints-grid');
if (!grid) return;
grid.innerHTML = JOINT_NAMES.map((name, i) => `
<div class="motor-row" id="mi-${i}">
<span class="stale" id="ms-${i}" style="color:#ef4444">●</span>
<span class="name">${name}</span>
<span class="val pos" id="mp-${i}">0.00</span>
<span class="val vel" id="mv-${i}">0.00</span>
<span class="val tau" id="mt-${i}">0.00</span>
</div>`).join('');
}
function updateJointsGrid(robot) {
if (!robot) return;
const pos = robot.joint_pos || [];
const vel = robot.joint_vel || [];
const tau = robot.joint_torque || [];
const upd = robot.update_counts || [];
for (let i = 0; i < 16; i++) {
const dot = $('ms-' + i);
const cnt = upd[i] ?? 0;
if (dot) dot.style.color = cnt > 0 ? '#30d158' : '#ef4444';
const p = $('mp-' + i); if (p) p.textContent = (pos[i] || 0).toFixed(2);
const v = $('mv-' + i); if (v) v.textContent = (vel[i] || 0).toFixed(2);
const t = $('mt-' + i);
if (t) {
t.textContent = (tau[i] || 0).toFixed(2);
t.style.color = Math.abs(tau[i] || 0) > 16 ? '#ff453a' : '#ff9f0a';
}
}
}
// ── State polling ─────────────────────────────────────────────────────────────
function setText(id, text, cls) {
const el = $(id);
if (!el) return;
el.textContent = text;
if (cls !== undefined) el.className = 'diag-value ' + cls;
}
function applyState(data) {
const connected = data.connected &&
(!data.local_receive_time || Date.now() / 1000 - data.local_receive_time < 2.5);
const stage = $('stage');
if (stage) {
if (!connected) {
stage.textContent = 'DISCONNECTED';
stage.className = 'stage DISCONNECTED';
return;
}
}
const mode = data.mode || 'UNKNOWN';
if (mode !== currentMode) highlightMode(mode);
const rt = data.runtime || {};
const src = rt.target_source || '--';
const srcCls = src === 'safety_brake' ? 'bad'
: src === 'runtime_policy' ? 'ok'
: src === 'runtime_zero_hold' ? 'active' : '';
setText('d-source', src, srcCls);
setText('d-released', String(rt.runtime_released ?? '--'),
rt.runtime_released ? 'ok' : '');
setText('d-alpha', (rt.release_alpha ?? '--') !== '--'
? Number(rt.release_alpha).toFixed(2) : '--');
setText('d-zero', String(rt.zero_command ?? '--'));
setText('d-estop', String(data.estop ?? '--'),
data.estop ? 'bad' : 'ok');
setText('d-mux', data.mux_status || '--');
const robot = data.robot || {};
const imuAge = robot.imu_age_ms ?? null;
setText('d-imu-fresh', String(robot.imu_fresh ?? '--'),
robot.imu_fresh ? 'ok' : 'bad');
setText('d-imu-age', imuAge !== null ? imuAge.toFixed(1) : '--',
imuAge !== null ? (imuAge > 200 ? 'bad' : imuAge > 60 ? 'warn' : 'ok') : '');
const grav = robot.projected_gravity;
setText('d-gravity', grav ? grav.map(v => Number(v).toFixed(2)).join(', ') : '--',
grav && grav[2] < -0.5 ? 'ok' : 'warn');
setText('d-holdover', String(robot.holdover_count ?? '--'),
(robot.holdover_count || 0) > 10 ? 'warn' : '');
const odomAge = robot.odom_age_ms ?? null;
setText('d-odom-age', odomAge !== null ? odomAge.toFixed(1) : '--',
odomAge !== null ? (odomAge > 500 ? 'bad' : odomAge > 200 ? 'warn' : 'ok') : '');
const lp = robot.odom_local_pos;
setText('d-odom-pos', lp ? `x=${Number(lp[0]).toFixed(2)} y=${Number(lp[1]).toFixed(2)}` : '--');
const cv = data.cmd_vel || {};
const lin = cv.linear || {};
const ang = cv.angular || {};
setText('cv-vx', (lin.x ?? 0).toFixed(3));
setText('cv-vy', (lin.y ?? 0).toFixed(3));
setText('cv-yaw', (ang.z ?? 0).toFixed(3));
updateJointsGrid(robot);
}
async function poll() {
try {
const res = await fetch('/api/state');
const data = await res.json();
applyState(data);
} catch (_) {}
}
// ── WEB mode heartbeat ────────────────────────────────────────────────────────
setInterval(() => {
if (currentMode === 'WEB' && !dragging) sendCmd();
}, 50);
// ── Event log ─────────────────────────────────────────────────────────────────
function appendEvent(kind, detail, cls) {
const el = $('events-log');
if (!el) return;
const div = document.createElement('div');
const t = new Date().toLocaleTimeString();
div.innerHTML = `<span class="ev-t">${t}</span> <span class="ev-${cls || 'ok'}">${kind}</span> <span style="color:#8e8e93">${detail || ''}</span>`;
el.appendChild(div);
while (el.children.length > 200) el.removeChild(el.firstChild);
el.scrollTop = el.scrollHeight;
}
// ── Init ──────────────────────────────────────────────────────────────────────
initJointsGrid();
setInterval(poll, 100);
appendEvent('READY', '页面已加载,等待 Nano 连接', 'ok');
@@ -0,0 +1,98 @@
<!doctype html>
<html lang="zh-CN" data-theme="dark">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width,initial-scale=1">
<title>sim2real ROS2 控制台</title>
<link rel="stylesheet" href="style.css">
</head>
<body>
<header class="glass-panel top-bar">
<div class="top-bar-left">
<h1>sim2real ROS2</h1>
<span class="stage" id="stage">DISCONNECTED</span>
</div>
<div class="top-bar-center">
<span class="label">控制模式</span>
<button class="btn" id="btn-disabled">DISABLED</button>
<button class="btn btn-remote" id="btn-remote">REMOTE</button>
<button class="btn btn-web" id="btn-web">WEB</button>
<button class="btn btn-nav" id="btn-nav">NAV</button>
<div class="divider"></div>
<button class="btn" id="btn-zero">速度归零</button>
</div>
<div class="top-bar-right">
<button class="btn btn-danger" id="btn-estop">软急停</button>
</div>
</header>
<div class="glass-panel side-panel left-panel">
<div class="panel-section">
<h2 class="panel-title">运行状态</h2>
<div class="diag-row"><span class="diag-label">target_source</span><span class="diag-value" id="d-source">--</span></div>
<div class="diag-row"><span class="diag-label">runtime_released</span><span class="diag-value" id="d-released">--</span></div>
<div class="diag-row"><span class="diag-label">release_alpha</span><span class="diag-value" id="d-alpha">--</span></div>
<div class="diag-row"><span class="diag-label">zero_command</span><span class="diag-value" id="d-zero">--</span></div>
<div class="diag-row"><span class="diag-label">estop</span><span class="diag-value" id="d-estop">--</span></div>
<div class="diag-row"><span class="diag-label">mux</span><span class="diag-value" id="d-mux">--</span></div>
</div>
<div class="panel-section">
<h2 class="panel-title">IMU &amp; 里程计</h2>
<div class="diag-row"><span class="diag-label">IMU fresh</span><span class="diag-value" id="d-imu-fresh">--</span></div>
<div class="diag-row"><span class="diag-label">IMU age ms</span><span class="diag-value" id="d-imu-age">--</span></div>
<div class="diag-row"><span class="diag-label">projected_gravity</span><span class="diag-value" id="d-gravity">--</span></div>
<div class="diag-row"><span class="diag-label">holdover</span><span class="diag-value" id="d-holdover">--</span></div>
<div class="diag-row"><span class="diag-label">odom age ms</span><span class="diag-value" id="d-odom-age">--</span></div>
<div class="diag-row"><span class="diag-label">odom local pos</span><span class="diag-value" id="d-odom-pos">--</span></div>
</div>
<div class="panel-section flex-1">
<h2 class="panel-title">关节状态 (16轴)</h2>
<div id="joints-grid" class="motors-grid-list"></div>
</div>
</div>
<div class="glass-panel side-panel right-panel">
<div class="panel-section">
<h2 class="panel-title">Web 手动控制</h2>
<div class="joystick-area">
<div class="joystick" id="joystick"><div id="stick"></div></div>
<p class="hint">拖动控制前后(vx)和横移(vy),松开归零</p>
</div>
<div class="slider-group">
<div class="slider-row">
<span class="slider-label">vx</span>
<input type="range" id="cmd-vx" class="glass-slider" min="-0.8" max="0.8" step="0.01" value="0">
<span class="slider-val" id="cmd-vx-v">0.00</span>
</div>
<div class="slider-row">
<span class="slider-label">vy</span>
<input type="range" id="cmd-vy" class="glass-slider" min="-0.3" max="0.3" step="0.01" value="0">
<span class="slider-val" id="cmd-vy-v">0.00</span>
</div>
<div class="slider-row">
<span class="slider-label">yaw</span>
<input type="range" id="cmd-yaw" class="glass-slider" min="-0.5" max="0.5" step="0.01" value="0">
<span class="slider-val" id="cmd-yaw-v">0.00</span>
</div>
</div>
<div class="cmd-display" id="cmd-display">vx=0.00 vy=0.00 yaw=0.00</div>
</div>
<div class="panel-section">
<h2 class="panel-title">当前输出 /cmd_vel</h2>
<div class="diag-row"><span class="diag-label">linear.x</span><span class="diag-value" id="cv-vx">--</span></div>
<div class="diag-row"><span class="diag-label">linear.y</span><span class="diag-value" id="cv-vy">--</span></div>
<div class="diag-row"><span class="diag-label">angular.z</span><span class="diag-value" id="cv-yaw">--</span></div>
</div>
<div class="panel-section log-section flex-1">
<h2 class="panel-title">事件流</h2>
<div id="events-log" class="log"></div>
</div>
</div>
<script src="app.js"></script>
</body>
</html>
@@ -0,0 +1,263 @@
/* sim2real ROS2 Web Debug — Apple Glass Design */
:root {
--bg-primary: #000000;
--glass-bg: rgba(20, 20, 22, 0.65);
--glass-border: rgba(255, 255, 255, 0.12);
--glass-shadow: 0 8px 32px rgba(0, 0, 0, 0.25);
--text-primary: #ffffff;
--text-secondary: #ebebf5;
--text-tertiary: #8e8e93;
--accent: #0a84ff;
--accent-hover: #409cff;
--success: #30d158;
--warning: #ffd60a;
--danger: #ff453a;
--blur-amount: 24px;
--saturation: 180%;
--spring: cubic-bezier(0.4, 0, 0.2, 1);
--panel-radius: 16px;
--font-family: -apple-system, BlinkMacSystemFont, "SF Pro Display", "PingFang SC", sans-serif;
}
* { box-sizing: border-box; margin: 0; padding: 0; }
body {
font-family: var(--font-family);
background: radial-gradient(circle at top left, #1a1a24 0%, #000000 100%);
color: var(--text-primary);
-webkit-font-smoothing: antialiased;
min-height: 100vh;
overflow-x: hidden;
}
.glass-panel {
background: var(--glass-bg);
backdrop-filter: blur(var(--blur-amount)) saturate(var(--saturation));
-webkit-backdrop-filter: blur(var(--blur-amount)) saturate(var(--saturation));
border: 0.5px solid var(--glass-border);
box-shadow: var(--glass-shadow);
z-index: 50;
}
/* Top Bar */
.top-bar {
position: fixed;
top: 16px;
left: 50%;
transform: translateX(-50%);
display: flex;
align-items: center;
justify-content: space-between;
padding: 8px 16px;
border-radius: 24px;
width: 96%;
max-width: 1400px;
gap: 16px;
}
.top-bar-left, .top-bar-center, .top-bar-right {
display: flex;
align-items: center;
gap: 10px;
}
.top-bar-center { flex: 1; justify-content: center; }
.top-bar h1 {
font-size: 16px;
font-weight: 600;
background: linear-gradient(45deg, #fff, #8e8e93);
-webkit-background-clip: text;
-webkit-text-fill-color: transparent;
}
.divider { width: 1px; height: 24px; background: var(--glass-border); margin: 0 4px; }
/* Stage badge */
.stage {
padding: 4px 10px;
border-radius: 12px;
font-size: 11px;
font-weight: 700;
text-transform: uppercase;
letter-spacing: 0.5px;
background: rgba(255,255,255,0.1);
color: var(--text-secondary);
}
.stage.DISCONNECTED { background: rgba(142,142,147,0.3); color: #aaa; }
.stage.CONNECTED { background: rgba(10,132,255,0.3); color: #82c4ff; }
.stage.REMOTE { background: rgba(48,209,88,0.3); color: #8deda7; }
.stage.WEB { background: rgba(0,122,255,0.3); color: #82c4ff; }
.stage.NAV { background: rgba(255,214,10,0.3); color: #ffe680; }
.stage.DISABLED { background: rgba(142,142,147,0.25); color: #aaa; }
.stage.ESTOPPED { background: rgba(255,69,58,0.5); color: #ff8b86; box-shadow: 0 0 8px rgba(255,69,58,0.4); }
/* Buttons */
.btn {
background: rgba(255,255,255,0.08);
border: 1px solid rgba(255,255,255,0.1);
border-radius: 8px;
color: var(--text-primary);
font-size: 12px;
font-weight: 500;
padding: 6px 12px;
cursor: pointer;
transition: all 0.2s var(--spring);
font-family: inherit;
}
.btn:hover:not(:disabled) { background: rgba(255,255,255,0.15); transform: translateY(-1px); }
.btn:active:not(:disabled) { transform: translateY(1px); }
.btn:disabled { opacity: 0.5; cursor: not-allowed; }
.btn-danger { background: rgba(255,69,58,0.8); border-color: transparent; color: white; }
.btn-remote { background: rgba(48,209,88,0.2); border-color: rgba(48,209,88,0.4); color: #8deda7; }
.btn-web { background: rgba(10,132,255,0.2); border-color: rgba(10,132,255,0.4); color: #82c4ff; }
.btn-nav { background: rgba(255,214,10,0.2); border-color: rgba(255,214,10,0.4); color: #ffe680; }
.btn.active-mode { box-shadow: 0 0 0 2px white; }
.label { font-size: 11px; color: var(--text-tertiary); }
/* Side panels */
.side-panel {
position: fixed;
top: 80px;
bottom: 20px;
width: 340px;
border-radius: var(--panel-radius);
display: flex;
flex-direction: column;
overflow: hidden;
}
.left-panel { left: 2%; }
.right-panel { right: 2%; }
.flex-1 { flex: 1; min-height: 0; overflow: hidden; display: flex; flex-direction: column; }
.panel-section {
padding: 14px 16px;
border-bottom: 0.5px solid var(--glass-border);
}
.panel-section:last-child { border-bottom: none; }
.panel-title {
font-size: 11px;
font-weight: 700;
color: var(--text-tertiary);
text-transform: uppercase;
letter-spacing: 0.5px;
margin-bottom: 10px;
}
/* Diag rows */
.diag-row {
display: flex;
justify-content: space-between;
align-items: center;
padding: 4px 6px;
background: rgba(0,0,0,0.2);
border-radius: 4px;
margin-bottom: 3px;
}
.diag-label { font-size: 11px; color: var(--text-tertiary); }
.diag-value { font-size: 11px; font-family: monospace; color: var(--text-primary); font-weight: 600; }
.diag-value.ok { color: var(--success); }
.diag-value.warn { color: var(--warning); }
.diag-value.bad { color: var(--danger); }
.diag-value.active { color: #82c4ff; }
/* Joints grid */
.motors-grid-list {
display: flex;
flex-direction: column;
gap: 2px;
overflow-y: auto;
flex: 1;
}
.motor-row {
display: flex;
align-items: center;
gap: 4px;
padding: 3px 6px;
background: rgba(0,0,0,0.25);
border-radius: 5px;
}
.motor-row .name { font-size: 10px; color: var(--text-secondary); width: 72px; font-family: monospace; flex-shrink: 0; }
.motor-row .val { font-size: 10px; font-family: monospace; text-align: right; flex: 1; }
.motor-row .val.pos { color: #0a84ff; }
.motor-row .val.vel { color: #30d158; }
.motor-row .val.tau { color: #ff9f0a; }
.motor-row .stale { font-size: 9px; width: 8px; flex-shrink: 0; }
/* Joystick */
.joystick-area { display: flex; flex-direction: column; align-items: center; gap: 8px; margin-bottom: 12px; }
.joystick {
position: relative;
width: 180px;
height: 180px;
border-radius: 50%;
background: radial-gradient(circle, rgba(10,132,255,0.2), rgba(10,132,255,0.05));
border: 1px solid rgba(10,132,255,0.3);
touch-action: none;
flex-shrink: 0;
}
#stick {
position: absolute;
left: 50%; top: 50%;
width: 56px; height: 56px;
border-radius: 50%;
background: linear-gradient(135deg, #30d158, #0a84ff);
transform: translate(-50%, -50%);
box-shadow: 0 8px 24px rgba(0,0,0,0.4);
}
/* Sliders */
.slider-group { display: flex; flex-direction: column; gap: 8px; }
.slider-row { display: flex; align-items: center; gap: 8px; }
.slider-label { font-size: 11px; color: var(--text-tertiary); width: 28px; font-family: monospace; }
.slider-val { font-size: 11px; color: var(--accent); font-family: monospace; width: 38px; text-align: right; }
.glass-slider {
flex: 1;
-webkit-appearance: none;
height: 4px;
border-radius: 2px;
background: rgba(255,255,255,0.2);
outline: none;
}
.glass-slider::-webkit-slider-thumb {
-webkit-appearance: none;
width: 14px; height: 14px;
border-radius: 50%;
background: white;
cursor: pointer;
box-shadow: 0 2px 4px rgba(0,0,0,0.5);
}
.cmd-display {
margin-top: 8px;
padding: 6px 10px;
background: rgba(0,0,0,0.3);
border-radius: 6px;
font-family: monospace;
font-size: 12px;
color: var(--accent);
text-align: center;
}
.hint { font-size: 11px; color: var(--text-tertiary); text-align: center; }
/* Log */
.log-section { flex: 1; overflow: hidden; display: flex; flex-direction: column; }
.log {
flex: 1;
background: rgba(0,0,0,0.4);
border: 1px solid rgba(255,255,255,0.05);
border-radius: 6px;
padding: 8px;
font-family: monospace;
font-size: 11px;
color: var(--text-secondary);
overflow-y: auto;
}
.log div { margin-bottom: 2px; line-height: 1.4; }
.ev-t { color: var(--text-tertiary); margin-right: 4px; }
.ev-ok { color: var(--success); }
.ev-warn { color: var(--warning); }
.ev-bad { color: var(--danger); }
@media (max-width: 960px) {
.side-panel { position: relative; top: auto; width: 100%; left: 0; right: 0; border-radius: 0; height: auto; }
.top-bar { width: 100%; border-radius: 0; top: 0; }
body { padding-top: 56px; }
}
@@ -0,0 +1,6 @@
build/
install/
log/
.colcon/
.vscode/
compile_commands.json
@@ -0,0 +1,178 @@
# ROS2 C++ Sim2Real 运动控制栈 - 部署指南
本工作区提供了一个自包含、独立的 C++ ROS2 Humble 实现,用于在 Jetson Orin 目标机上部署轮腿四足机器人控制策略。
---
## 1. 前提条件与环境
### 硬件
* **目标计算机**:运行 Ubuntu 22.04 LTS 的 Jetson Orin Nano / Orin NX / AGX Orin。
* **IMU 传感器**Odin 集成 IMU,发布至 `/odin1/imu`
* **CAN 总线适配器**Peak CAN、USB-to-CAN 或板载 SocketCAN 接口,使用 CAN0 和 CAN1。
### 主机依赖
* **操作系统**Ubuntu 22.04 LTS (Jammy Jellyfish)。
* **ROS 2 发行版**ROS 2 HumbleDesktop-Base 或 ROS-Base)。
* **C++ 编译器**:支持 C++17 的 GCC/G++ 9.0+。
* **库与 ROS2 包**
* `libyaml-cpp-dev`
* `libeigen3-dev`
* `libusb-1.0-0-dev`Odin USB 传感器通信)
* `libpcl-dev``libopencv-dev`3D 点云与相机处理)
* `ros-humble-navigation2``ros-humble-nav2-bringup`Nav2 规划器/控制器服务器)
* `ros-humble-pointcloud-to-laserscan`(点云转激光扫描,供 AMCL 使用)
* `ros-humble-cv-bridge``ros-humble-pcl-conversions`(Odin 传感器驱动图像与点云处理)
* `can-utils`SocketCAN 验证工具)
---
## 2. 本地编译与部署
按以下步骤在主机系统上编译运行整个栈:
### 步骤 1:安装系统依赖
```bash
sudo apt-get update
sudo apt-get install -y build-essential cmake can-utils libyaml-cpp-dev libeigen3-dev \
libusb-1.0-0-dev libpcl-dev libopencv-dev ros-humble-navigation2 \
ros-humble-nav2-bringup ros-humble-pointcloud-to-laserscan \
ros-humble-cv-bridge ros-humble-pcl-conversions
```
### 步骤 2:下载 ONNXRuntime C++ SDK
策略需要 ONNXRuntime 库来运行推理。必须下载并解压到已知目录:
```bash
# 创建目录
sudo mkdir -p /opt/onnxruntime
cd /opt
# 针对 Jetson Orin (ARM64 / aarch64)
sudo wget https://github.com/microsoft/onnxruntime/releases/download/v1.16.3/onnxruntime-linux-aarch64-1.16.3.tgz
sudo tar -zxvf onnxruntime-linux-aarch64-1.16.3.tgz --strip-components=1 -C /opt/onnxruntime
# 或标准桌面仿真 (x86_64 / amd64)
# sudo wget https://github.com/microsoft/onnxruntime/releases/download/v1.16.3/onnxruntime-linux-x64-1.16.3.tgz
# sudo tar -zxvf onnxruntime-linux-x64-1.16.3.tgz --strip-components=1 -C /opt/onnxruntime
```
导出 CMake 辅助变量:
```bash
export ONNXRUNTIME_DIR=/opt/onnxruntime
```
### 步骤 3:构建工作区
进入包含 `src/` 的本包根目录,运行 `colcon`
```bash
colcon build --merge-install --cmake-args -DCMAKE_BUILD_TYPE=Release
```
### 步骤 4:配置 SocketCAN 接口
启动前,以 1 Mbps 波特率激活 CAN 接口:
```bash
sudo ip link set can0 up type can bitrate 1000000
sudo ip link set can1 up type can bitrate 1000000
```
使用 `ifconfig``ip link` 验证接口已启动。
### 步骤 5:启动节点
使启动脚本可执行并运行:
```bash
chmod +x start_sim2real.sh
./start_sim2real.sh
```
---
## 3. Docker 部署(推荐)
强烈推荐使用 Docker 隔离依赖,避免 Jetson Orin 上的库版本冲突。
### 步骤 1:构建镜像
确保在 `sim2real_ros2` 目录中(包含 `Dockerfile`):
```bash
# 使用标准 docker build
docker build -t sim2real_ros2:latest .
# 或使用 Docker Compose
docker compose build
```
### 步骤 2:运行容器
对于真实硬件部署,容器**必须**共享主机网络栈(用于 ROS2 DDS 和 SocketCAN)并具备线程优先级能力以实现实时调度:
```bash
# 选项 A:手动运行
docker run -it \
--network host \
--privileged \
--cap-add=sys_nice \
--volume=/dev:/dev \
--shm-size=2g \
--name sim2real_ros2_run \
sim2real_ros2:latest
# 选项 B:通过 Docker Compose 运行(最简单)
docker compose up -d
```
---
## 4. 系统拓扑与话题
控制节点通过标准 ROS 2 DDS 消息与传感器驱动和导航栈交互:
* **IMU 输入**:订阅 `/odin1/imu``sensor_msgs/msg/Imu`)。硬件节点自动执行逆轴旋转(`x_raw = -y_ros``y_raw = x_ros`)以重建 RL 策略期望的原始坐标系。
* **控制命令**:订阅 `/cmd_vel``/cmd_vel_stamped``geometry_msgs/msg/Twist` / `TwistStamped`),由导航栈或手动键盘节点发布。
* **里程计输入**:订阅 `/odom``nav_msgs/msg/Odometry`),由 `odom_relay_node``/odin1/odometry` 中继并重映射帧名后提供。
* **急停**:订阅 `/safety/estop``std_msgs/msg/Bool`)。发布 `true` 触发软件急停,机器人进入低刚度阻尼刹车。
* **状态遥测**:发布 `runtime/state``sim2real_interfaces/msg/RuntimeState`),包含当前关节速度、温度、IMU 输出和诊断信息。
* **策略目标**:发布 `runtime/target``sim2real_interfaces/msg/RuntimeTarget`),包含策略推理输出的目标关节位置。
### TF 树
```
odom ──→ base_link (由 odom_relay_node 广播)
map ──→ odom (由 AMCL / Odin SLAM 发布,取决于运行模式)
```
---
## 5. 集成 ROS 2 导航与传感器驱动
### USB 设备权限(Odin 传感器)
要运行物理 Odin 传感器驱动(`odin_ros_driver`),目标计算机必须具有传感器 USB 接口的读写权限。在主机系统上添加以下 udev 规则:
```bash
# 1. 添加 udev 规则
echo 'SUBSYSTEM=="usb", ATTR{idVendor}=="2207", ATTR{idProduct}=="0019", MODE="0666", GROUP="plugdev"' | sudo tee /etc/udev/rules.d/99-odin-usb.rules
# 2. 重新加载 udev 规则并重新插拔传感器
sudo udevadm control --reload
sudo udevadm trigger
```
### 集成启动参数
统一启动文件 `sim2real_system.launch.py` 支持模块化激活传感器驱动和 Nav2 导航栈:
* `launch_driver`(默认:`true`):启动 `odin_ros_driver` 节点以获取 IMU 和点云遥测。
* `launch_nav2`(默认:`true`):启动 ROS2 Navigation2 规划器、控制器、costmap、AMCL 和 pointcloud_to_laserscan。
#### 1. 完整真实硬件闭环(默认)
启动运动控制运行时、物理 CAN 桥接、Odin 传感器驱动和 Nav2 导航:
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py dry_run:=false launch_driver:=true launch_nav2:=true
```
#### 2. Dry-Run / 仿真航点测试
在 dry-run 模式下运行策略运行时和 Nav2 导航(不访问 CAN 总线或物理 USB 传感器,适合测试导航话题路由):
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py dry_run:=true launch_driver:=false launch_nav2:=true
```
#### 3. 仅运动控制(无导航)
禁用传感器驱动和 Nav2,让运动策略等待 `/cmd_vel` 上的手动速度输入(如键盘遥操作):
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_driver:=false launch_nav2:=false
```
@@ -0,0 +1,80 @@
# 使用 ROS2 官方 Humble 基础镜像
FROM ros:humble-ros-base-jammy
ENV DEBIAN_FRONTEND=noninteractive
# 安装 C++ 编译依赖、SocketCAN 调试工具及 Eigen 等核心库
RUN apt-get update && apt-get install -y --no-install-recommends \
build-essential \
cmake \
git \
can-utils \
libyaml-cpp-dev \
libeigen3-dev \
libusb-1.0-0-dev \
libpcl-dev \
libopencv-dev \
ros-humble-navigation2 \
ros-humble-nav2-bringup \
ros-humble-pointcloud-to-laserscan \
ros-humble-cv-bridge \
ros-humble-pcl-conversions \
wget \
tar \
python3-pip \
&& rm -rf /var/lib/apt/lists/*
# ============================================================================
# ONNX Runtime — 架构自适应,aarch64 启用 CUDA GPU 加速
# ============================================================================
# - Orin Nano (aarch64): pip 安装 onnxruntime-gpu(含 CUDA EP
# - x86_64 开发机: 下载 CPU-only 预编译包(GPU 不可用)
WORKDIR /opt
RUN ARCH=$(uname -m) && \
if [ "$ARCH" = "aarch64" ]; then \
echo "[ONNX] Installing CUDA-enabled ONNX Runtime for Jetson Orin..." && \
pip3 install --no-cache-dir onnxruntime-gpu && \
SITE_PKGS=$(python3 -c "import site; print(site.getsitepackages()[0])") && \
mkdir -p onnxruntime/include onnxruntime/lib && \
cp -r "$SITE_PKGS/onnxruntime/include/"* onnxruntime/include/ && \
cp "$SITE_PKGS/onnxruntime/capi/libonnxruntime.so"* onnxruntime/lib/ && \
echo "[ONNX] CUDA ONNX Runtime installed."; \
else \
echo "[ONNX] Installing CPU-only ONNX Runtime for x86_64 dev..." && \
wget -q https://github.com/microsoft/onnxruntime/releases/download/v1.16.3/onnxruntime-linux-x64-1.16.3.tgz && \
tar -zxf onnxruntime-linux-x64-1.16.3.tgz && \
mv onnxruntime-linux-x64-1.16.3 onnxruntime && \
rm onnxruntime-linux-x64-1.16.3.tgz; \
fi
ENV ONNXRUNTIME_DIR=/opt/onnxruntime
# 创建工作空间,将所有 C++ 源码包拷入
WORKDIR /sim2real_ws/src
COPY src/sim2real_bringup sim2real_bringup
COPY src/sim2real_common sim2real_common
COPY src/sim2real_hw sim2real_hw
COPY src/sim2real_interfaces sim2real_interfaces
COPY src/sim2real_runtime sim2real_runtime
COPY src/odin_ros_driver odin_ros_driver
COPY src/sim2real_nav2 sim2real_nav2
# 拷贝策略文件与运行脚本
WORKDIR /sim2real_ws
COPY policies policies
COPY map map
COPY tools tools
COPY start_sim2real.sh start_sim2real.sh
RUN chmod +x start_sim2real.sh
# 编译 ROS2 工作空间
SHELL ["/bin/bash", "-c"]
RUN source /opt/ros/humble/setup.bash && \
colcon build --merge-install --cmake-args -DCMAKE_BUILD_TYPE=Release
# 拷贝 Docker 入口脚本并设置
COPY docker_entrypoint.sh /docker_entrypoint.sh
RUN chmod +x /docker_entrypoint.sh
ENTRYPOINT ["/docker_entrypoint.sh"]
CMD ["./start_sim2real.sh"]
@@ -0,0 +1,76 @@
# ROS 2 Sim2Real v2:里程计导航联调
本目录归档 `real/sim2real_ros2_v2(odom)`,对应重排主线的 `v0.12.0`。该阶段在 `v0.11.1` 的 Odin/TensorRT 与站姿调参基础上,固定纯里程计模式,增加 odom fallback 的 TF 冲突保护、A_min 路线和地图工具。
本工程保留当前 `sim2real` 已验证的部署契约,同时将运行时热路径迁移到 C++:
- `53D` 策略观测契约不变
- `16D` 动作契约不变
- `50Hz` 策略循环与训练对齐
- `200Hz` 电机循环为专用 C++ 热路径
- ROS 2 作为导航、TF、诊断和启动管理的系统集成层
## 工作区布局
- `src/sim2real_interfaces`
硬件桥接与策略运行时共享的 ROS 2 消息定义。
- `src/sim2real_common`
共享常量、部署契约辅助函数、Mahony 姿态滤波器、站立平衡控制器、安全监控。
- `src/sim2real_hw`
面向硬件的桥接节点:RobStride CAN 收发、IMU/Odin 数据采集、看门狗、状态发布。
- `src/sim2real_runtime`
策略运行时节点:`53D→16D` ONNX 推理、命令滤波/仲裁、目标发布。
同时包含 `odom_relay_node`(里程计中继与 TF 广播)。
- `src/sim2real_nav2`
ROS 2 Navigation2 (Nav2) 配置包:参数、启动文件、AMCL、costmap、planner/controller。
- `src/sim2real_bringup`
统一启动文件与运行时参数配置。
- `src/odin_ros_driver`
Odin 传感器 ROS 2 驱动(含 IMU、点云、里程计发布)。
- `docs`
架构说明与迁移计划。
## 目标架构
```text
Odin / IMU / Odom ---> sim2real_hw ---> sim2real_runtime ---> sim2real_hw
| | |
v v v
RuntimeState RuntimeTarget 电机 CAN 指令
| |
+-------> 诊断 / 遥测
Nav2 / cmd_vel ------------------------------> sim2real_runtime
(经 odom_relay_node 提供 odom→base_link TF)
```
## 当前状态
已完成 Phase 0-5 的全部迁移:
1. ✅ 冻结部署契约(deployment_contract.hpp
2. ✅ ROS 2 包结构搭建
3. ✅ 硬件热路径迁移至 C++SocketCAN 驱动、200Hz 电机循环)
4. ✅ ONNX 策略运行时迁移至 C++(50Hz 推理循环)
5. ✅ 导航与诊断通过 ROS 2 接入(Nav2 + odom_relay + TF
## 契约来源
迁移过程中以下文件被视为真值源:
- `sim2real/deployment_manifest.yaml`
- `sim2real/interface/motor_mapping.py`
- `sim2real/interface/real_io.py`
- `sim2real/policy/policy_runner.py`
- `sim2real/web/session.py`
## 注意事项
- 开发目标为 Linux + ROS 2 Humble,运行于 Jetson Orin / x86_64。
- Windows 仅作为编辑环境使用。
- 观测顺序、动作缩放、默认站姿、电机映射不得独立修改,
除非训练与部署同步更新。
- 默认 Rough 策略为 `model_9600`,默认站姿回到比赛站姿髋俯仰 `0.550`、膝关节 `-1.125`
- Odin `custom_map_mode` 固定为 `0`(纯里程计),避免没有 `map_a.bin` 时产生 map/odom TF 冲突。
- `A_min.pcd``C.pcd``map_b.pcd` 均为确定性抽样预览,点数和哈希见 [`map/README.md`](map/README.md)。
- TensorRT engine 与 JetPack、TensorRT 版本及 GPU 架构有关,其他机器应从 ONNX 重新生成。
@@ -0,0 +1,22 @@
version: '3.8'
services:
sim2real_ros2:
build:
context: .
dockerfile: Dockerfile
container_name: sim2real_ros2_node
runtime: nvidia
network_mode: host
privileged: true
stdin_open: true
tty: true
environment:
- NVIDIA_VISIBLE_DEVICES=all
- NVIDIA_DRIVER_CAPABILITIES=compute,utility
cap_add:
- SYS_NICE
shm_size: '2gb'
volumes:
- /dev:/dev
restart: unless-stopped
@@ -0,0 +1,12 @@
#!/bin/bash
set -e
# Source ROS2 Humble environment
source /opt/ros/humble/setup.bash
# Source workspace install setup if compiled
if [ -f "/sim2real_ws/install/setup.bash" ]; then
source /sim2real_ws/install/setup.bash
fi
exec "$@"
@@ -0,0 +1,102 @@
# sim2real_ros2 架构说明
## 设计目标
- 保留已验证的 RL 部署契约不变
- 将低延迟循环从 Python 迁移至 C++
- 暴露标准 ROS 2 接口用于导航和系统集成
- 保持安全边界独立于策略正确性
## 各包职责
### `sim2real_interfaces`(接口消息)
定义最小化的运行时消息:
- `RuntimeState`
硬件桥接发布的归一化运行时状态快照
- `RuntimeTarget`
策略运行时发送至硬件桥接的最新策略目标
### `sim2real_common`(共享常量)
存储编译期常量和部署契约辅助:
- 观测维度和字段布局
- 动作维度和轮子索引
- 关节顺序和默认站姿
- 动作缩放因子和默认循环频率
- Mahony 姿态滤波器
- 站立平衡控制器
- 安全监控器(SafetyMonitor / RuntimeGuard
### `sim2real_hw`(硬件桥接)
拥有硬件侧执行循环和安全边界:
- RobStride CAN 收发
- IMU 与 Odin 状态采集
- 电机丢帧检测与保活逻辑(holdover)
- 看门狗与阻尼刹车
- 发布 `RuntimeState`
- 订阅 `RuntimeTarget`
- 订阅 `/odom` 里程计数据
目标热路径:
- 以 `200Hz` 频率读取状态
- 应用最新安全目标
- 超时或安全违规时立即停机
### `sim2real_runtime`(策略运行时)
拥有策略侧执行:
- 订阅 `RuntimeState`
- 按当前部署契约精确构建 `53D` 观测
- 以 `50Hz` 运行 ONNXRuntime 推理
- 对 raw_action 做 `[-10, 10]` 安全裁剪
- 发布 `RuntimeTarget`
- 仲裁命令来源:estop > safety_hold > startup > navigation > web
同时包含:
- `odom_relay_node`:将 `/odin1/odometry` 中继为 `/odom`,帧名 `odin1_base_link``base_link`,并广播 TF
### `sim2real_nav2`(导航配置)
拥有:
- Nav2 参数文件(planner、controller、costmap、AMCL、behavior
- Nav2 启动文件(含 AMCL、costmap 生命周期节点、pointcloud_to_laserscan
### `sim2real_bringup`(启动管理)
拥有:
- 参数文件
- 启动组合
- 运行时模式选择
- 集成 odin_ros_driver、sim2real_nav2 的条件启动
## 迁移规则
1. 优化之前先冻结当前契约
2. 先迁移传输和循环结构,再调整控制算法
3. C++ 运行时未达到影子模式一致性前,保留 Python 运行时可用
4. 按段测量延迟:
- 观测延迟
- 策略推理延迟
- 目标传输延迟
- 执行器响应延迟
## 首个里程碑
首个里程碑不是"机器人在 ROS 2 下行走",而是:
1. `sim2real_hw` 发布稳定的 `RuntimeState`
2. `sim2real_runtime` 从该状态构建正确的 `53D` 观测
3. `sim2real_runtime``50Hz` 发布 `RuntimeTarget`
4. `sim2real_hw` 消费最新目标并执行超时刹车
5. `cmd_vel` 可通过 ROS 2 注入而不改变策略契约
> ✅ 以上里程碑已全部完成。
@@ -0,0 +1,80 @@
# 迁移计划
## Phase 1: 硬件核心迁移 ✅ 已完成
将当前高频热路径从 Python 迁出。
吸收的源文件:
- `sim2real/interface/motor_driver.py`
- `sim2real/interface/motor_mapping.py`
- `sim2real/interface/imu_client.py`
- `sim2real/safety/runtime_guard.py`
- `sim2real/web/session.py`
交付物:
- C++ SocketCAN 电机总线封装
- C++ 状态缓存
- target 超时保活(timeout hold
- 阻尼刹车 / 急停通路
- 发布 `RuntimeState`
## Phase 2: 策略运行时迁移 ✅ 已完成
吸收的源文件:
- `sim2real/policy/policy_runner.py`
- `sim2real/interface/real_io.py`
- `sim2real/web/session.py`
交付物:
- 精确的 `53D` 观测构造器
- ONNXRuntime C++ 推理封装
- `50Hz` 策略定时器
- 命令平滑与来源仲裁
- raw_action `[-10, 10]` 安全裁剪
- 发布 `RuntimeTarget`
## Phase 3: ROS 2 系统集成 ✅ 已完成
参考的源项目:
- `00_ reference/odin_ros_driver`
- `00_ reference/EDULITE_A3/el_a3_ros`
- `00_ reference/rl_sar`
交付物:
- `cmd_vel` / `cmd_vel_stamped` 输入(支持 Twist 和 TwistStamped
- `odom_relay_node`:里程计中继 + TF 广播(odom → base_link
- 诊断话题
- rosbag/foxglove 可观测性
## Phase 4: 导航集成 ✅ 已完成
目标:
- 导航通过 ROS 2 发送身体速度指令
- RL 运行时保持为 locomotion 控制器
- 看门狗和安全边界始终在导航之下
规则:
- 导航绝不直接写电机指令
- 策略契约在重新训练前保持不变
- 任何新增历史项或里程计项必须版本化
## 当前状态
所有 4 个 Phase 已全部完成。以下为已实现的关键组件:
| 组件 | 节点 | 说明 |
|------|------|------|
| 硬件桥接 | `sim2real_hw_node` | 200Hz CAN 收发 + IMU + Mahony + 安全 |
| 策略运行时 | `sim2real_runtime_node` | 50Hz ONNX 推理 + 53D 观测 + raw_action clip |
| 里程计中继 | `odom_relay_node` | /odin1/odometry → /odom + odom→base_link TF |
| 导航栈 | Nav2 全套节点 | AMCL + costmap + DWB + Navfn + BT + lifecycle |
| 传感器驱动 | `odin_ros_driver` | IMU + 点云 + 里程计原始发布 |
| 点云转换 | `pointcloud_to_laserscan` | /odin1/cloud_slam → /scan (供 AMCL 使用) |
@@ -0,0 +1,495 @@
# sim2real_ros2 遥控器调用说明
本文档说明如何在 `sim2real_ros2` 中调用已接入的 SBUS UART 遥控器节点,以及执行后系统会产生什么效果。
## 1. 当前接入关系
遥控器节点位于:
```text
src/sim2real_runtime/src/remote_uart_node.py
```
该节点读取 SBUS 串口数据,并发布标准 ROS 2 控制话题:
| 输入 | 输出 | 作用 |
|---|---|---|
| SBUS UART 遥控器 | `/cmd_vel` | 给策略运行时发送速度命令 |
| SBUS CH7 高位 | `/safety/estop` | 触发软件急停 |
策略节点 `sim2real_runtime_node` 已经订阅 `/cmd_vel``/safety/estop`,所以遥控器不直接控制电机,而是通过 ROS 2 标准速度接口进入策略控制链路。
## 2. 通道映射
通道映射与前一阶段 Python Sim2Real 中的遥控器实现保持一致。
| 遥控器通道 | ROS 2 输出 | 含义 | 默认最大值 |
|---|---|---|---:|
| `CH2` | `cmd_vel.linear.x` | 前后速度 `vx` | `0.8 m/s` |
| `CH4` | `cmd_vel.linear.y` | 左右速度 `vy` | `0.3 m/s` |
| `CH1` | `cmd_vel.angular.z` | 转向角速度 `yaw` | `0.5 rad/s` |
| `CH7 HIGH` | `/safety/estop = true` | 软件急停 | - |
默认方向反转配置:
| 参数 | 默认值 | 含义 |
|---|---:|---|
| `remote_invert_vx` | `true` | 反转前后方向 |
| `remote_invert_vy` | `false` | 不反转横移方向 |
| `remote_invert_yaw` | `true` | 反转转向方向 |
## 3. 参数位置
遥控器参数在:
```text
src/sim2real_bringup/config/runtime.yaml
```
当前默认参数:
```yaml
remote_enabled: true
remote_port: "/dev/ttyACM0"
remote_baudrate: 100000
remote_timeout: 0.02
remote_axis_deadzone: 50
remote_active_threshold: 50
remote_axis_full_scale: 660.0
remote_max_vx: 0.8
remote_max_vy: 0.3
remote_max_yaw_rate: 0.5
remote_invert_vx: true
remote_invert_vy: false
remote_invert_yaw: true
remote_publish_inactive_zero: true
remote_estop_latch: true
remote_poll_hz: 50.0
```
如果遥控器串口不是 `/dev/ttyACM0`,需要修改:
```yaml
remote_port: "/dev/ttyUSB0"
```
或改成实际设备路径。
## 4. 启动前检查
### 4.1 确认串口存在
```bash
ls /dev/ttyACM* /dev/ttyUSB*
```
如果使用默认配置,应能看到:
```bash
/dev/ttyACM0
```
### 4.2 确认串口权限
如果节点提示串口权限不足,可以临时执行:
```bash
sudo chmod 666 /dev/ttyACM0
```
更推荐的长期方式是把当前用户加入 `dialout` 组:
```bash
sudo usermod -aG dialout $USER
```
然后重新登录。
### 4.3 确认 Python serial 依赖
节点依赖 `pyserial`。如果系统没有安装:
```bash
sudo apt update
sudo apt install -y python3-serial
```
## 5. 构建
如果刚修改过代码或参数,建议重新构建相关包:
```bash
cd /path/to/sim2real_ros2_v2
source /opt/ros/humble/setup.bash
colcon build --packages-select sim2real_runtime sim2real_bringup --symlink-install --merge-install
```
构建完成后 source 环境:
```bash
source install/setup.bash
```
确认可执行节点存在:
```bash
ros2 pkg executables sim2real_runtime
```
应包含:
```text
sim2real_runtime remote_uart_node.py
```
## 6. 推荐启动方式
### 6.1 启动完整系统,不启动 Nav2
这是你当前常用方式:
```bash
cd /path/to/sim2real_ros2_v2
source /opt/ros/humble/setup.bash
source install/setup.bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false
```
默认情况下,`launch_remote:=true`,所以上面命令会同时启动遥控器节点。
等价完整写法:
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false launch_remote:=true
```
### 6.2 不启动遥控器
如果只想用手动 `ros2 topic pub` 或其他上位机发 `/cmd_vel`,可以关闭遥控器节点:
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false launch_remote:=false
```
## 7. 单独启动遥控器节点
如果系统已经在运行,只想单独测试遥控器节点:
```bash
cd /path/to/sim2real_ros2_v2
source /opt/ros/humble/setup.bash
source install/setup.bash
ros2 run sim2real_runtime remote_uart_node.py --ros-args --params-file src/sim2real_bringup/config/runtime.yaml
```
如果要临时指定串口:
```bash
ros2 run sim2real_runtime remote_uart_node.py --ros-args \
--params-file src/sim2real_bringup/config/runtime.yaml \
-p remote_port:=/dev/ttyUSB0
```
## 8. 执行后会产生什么效果
启动以下命令后:
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false
```
系统会产生以下效果。
### 8.1 启动硬件桥接节点
节点:
```text
/sim2real_hw_node
```
效果:
1. 打开 `can0``can1`
2. 如果 `dry_run: false` 且 CAN 初始化成功,会使能 16 个 RobStride 电机。
3. 设置电机 MIT 模式。
4. 设置电机速度限制和力矩限制。
5. 以 `200Hz` 运行硬件读写循环。
6. 发布 `/runtime/state`
7. 订阅 `/runtime/target` 执行策略目标。
### 8.2 启动策略运行节点
节点:
```text
/sim2real_runtime_node
```
效果:
1. 加载 ONNX 策略模型。
2. 订阅 `/runtime/state`
3. 订阅 `/cmd_vel`
4. 订阅 `/safety/estop`
5. 执行启动站立流程:
- `boot_hold`
- `startup_soft_hold`
- `startup_hold`
- `runtime_zero_hold`
- `runtime_policy`
6. 以 `50Hz` 发布 `/runtime/target`
### 8.3 启动遥控器节点
节点:
```text
/sim2real_remote_uart_node
```
效果:
1. 打开默认串口 `/dev/ttyACM0`
2. 以 `50Hz` 轮询 SBUS 数据。
3. 遥控器摇杆居中时持续发布零速度:
```text
/cmd_vel:
linear.x = 0.0
linear.y = 0.0
angular.z = 0.0
```
4. 推动遥控器时发布非零速度,例如:
```text
/cmd_vel:
linear.x = vx
linear.y = vy
angular.z = yaw
```
5. 当 CH7 打到高位时发布:
```text
/safety/estop: true
```
由于当前 `remote_estop_latch: true`,急停是锁存式行为:一旦 CH7 高位触发,节点会发布急停,并保持内部急停已触发状态。恢复运行通常需要重启系统或手动发布复位信号,并确认机器人安全。
### 8.4 机器人行为效果
正常启动后,机器人不会立即按策略行走,而是按阶段执行:
1. 电机使能。
2. 读取当前关节位置。
3. 软保持当前姿态。
4. 平滑过渡到默认站立姿态。
5. 稳定后进入 runtime。
6. 遥控器无输入时保持站立平衡,即 `runtime_zero_hold`
7. 遥控器有输入时进入策略控制,即 `runtime_policy`
也就是说:
| 遥控器状态 | 机器人效果 |
|---|---|
| 摇杆居中 | 站立保持,不主动行走 |
| CH2 前后推动 | 前进/后退 |
| CH4 左右推动 | 横向移动 |
| CH1 左右推动 | 原地转向 |
| CH7 高位 | 软件急停,进入安全刹车 |
## 9. 如何确认遥控器已经生效
### 9.1 查看节点是否存在
```bash
ros2 node list
```
应看到:
```text
/sim2real_remote_uart_node
/sim2real_runtime_node
/sim2real_hw_node
```
### 9.2 查看 `/cmd_vel`
```bash
ros2 topic echo /cmd_vel
```
摇动遥控器时应看到 `linear.x``linear.y``angular.z` 变化。
### 9.3 查看 `/safety/estop`
```bash
ros2 topic echo /safety/estop
```
CH7 高位时应看到:
```yaml
data: true
```
### 9.4 查看策略目标阶段
```bash
ros2 topic echo /runtime/target --field target_source
```
常见输出含义:
| `target_source` | 含义 |
|---|---|
| `boot_hold` | 刚启动,保持初始姿态 |
| `startup_soft_hold` | 启动软保持 |
| `startup_hold` | 正在站立或站立后保持 |
| `runtime_zero_hold` | 已进入 runtime,遥控器无有效输入 |
| `runtime_policy` | 遥控器有输入,策略已经介入 |
| `safety_brake` | 安全刹车 |
| `timeout_hold` | 目标超时,硬件保持默认姿态 |
### 9.5 查看完整目标状态
```bash
ros2 topic echo --once /runtime/target
```
重点关注字段:
```yaml
target_source:
zero_command:
runtime_released:
release_alpha:
command:
raw_command:
```
如果遥控器摇杆有输入,通常会看到:
```yaml
target_source: runtime_policy
zero_command: false
runtime_released: true
release_alpha: 1.0
```
## 10. 常见问题
### 10.1 启动后提示无法打开串口
可能原因:
1. 串口路径不对。
2. 权限不足。
3. 设备没有插好。
4. 设备被其他程序占用。
检查:
```bash
ls /dev/ttyACM* /dev/ttyUSB*
```
修改 `runtime.yaml`
```yaml
remote_port: "/dev/ttyUSB0"
```
### 10.2 `/cmd_vel` 没有变化
检查:
```bash
ros2 node list
ros2 topic echo /cmd_vel
```
如果节点存在但无变化,可能是:
1. 遥控器没有输出 SBUS。
2. 串口波特率不对。
3. SBUS 接线错误。
4. 遥控器通道未校准。
5. 死区 `remote_axis_deadzone``remote_active_threshold` 太大。
### 10.3 摇杆方向反了
修改:
```yaml
remote_invert_vx: true
remote_invert_vy: false
remote_invert_yaw: true
```
例如前后方向反了,就切换:
```yaml
remote_invert_vx: false
```
### 10.4 急停后不恢复
当前配置:
```yaml
remote_estop_latch: true
```
这表示急停锁存。触发后建议:
1. 先确认机器人物理安全。
2. 停止 launch。
3. 将 CH7 打回安全位置。
4. 重新启动系统。
如果需要非锁存模式,可以改为:
```yaml
remote_estop_latch: false
```
但实机调试时更建议使用锁存模式。
## 11. 快速验证命令清单
```bash
cd /path/to/sim2real_ros2_v2
source /opt/ros/humble/setup.bash
source install/setup.bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false
```
另开终端:
```bash
cd /path/to/sim2real_ros2_v2
source /opt/ros/humble/setup.bash
source install/setup.bash
ros2 node list
ros2 topic echo /cmd_vel
ros2 topic echo /runtime/target --field target_source
```
如果只测遥控器,不启动电机系统:
```bash
ros2 run sim2real_runtime remote_uart_node.py --ros-args --params-file src/sim2real_bringup/config/runtime.yaml
```
另开终端:
```bash
ros2 topic echo /cmd_vel
ros2 topic echo /safety/estop
```
@@ -0,0 +1,246 @@
# sim2real_ros2 Web UDP 调试说明
本文档说明本次新增的最小 Web 调试链路。
## 1. 架构
```text
Windows 本地浏览器/HTTP 服务
|
| UDP JSON
v
Nano: sim2real_web_udp_bridge_node.py
|
| ROS 2 topics
v
sim2real_cmd_mux_node.py -> /cmd_vel -> sim2real_runtime_node
```
Web 页面在 Windows 本地渲染,Nano 只运行轻量 UDP bridge 和 ROS2 节点。
## 2. 新增 ROS2 节点
### `remote_uart_node.py`
遥控器节点现在发布:
```text
/cmd_vel_remote
```
不再直接发布 `/cmd_vel`
通道触发阈值改为:
```yaml
remote_axis_deadzone: 40
remote_active_threshold: 40
```
只有通道归一化值绝对值大于 `40` 才认为是有效输入。
### `cmd_mux_node.py`
输入:
```text
/cmd_vel_remote
/cmd_vel_web
/cmd_vel_nav
/control/mode
/remote/enabled
/web/enabled
/nav/enabled
/safety/estop
```
输出:
```text
/cmd_vel
/control/mode_state
/control/mux_status
```
控制模式:
```text
DISABLED
REMOTE
WEB
NAV
```
急停 `/safety/estop=true` 会强制进入 `DISABLED`,并输出零速度。
### `web_udp_bridge_node.py`
Nano 端 UDP 监听:
```text
0.0.0.0:15000
```
发布:
```text
/cmd_vel_web
/safety/estop
/control/mode
/web/enabled
/remote/enabled
/nav/enabled
```
订阅并回传状态:
```text
/runtime/state
/runtime/target
/cmd_vel
/safety/estop
/control/mode_state
/control/mux_status
```
## 3. Nano 启动
```bash
cd /path/to/sim2real_ros2_v2
source /opt/ros/humble/setup.bash
source install/setup.bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false
```
默认会启动:
```text
sim2real_remote_uart_node
sim2real_cmd_mux_node
sim2real_web_udp_bridge_node
```
如果不想启动 Web UDP bridge
```bash
ros2 launch sim2real_bringup sim2real_system.launch.py launch_nav2:=false launch_web_bridge:=false
```
## 4. Windows 本地 Web 启动
把目录复制到 Windows 或通过共享目录访问:
```text
tools/win_web_debug
```
在 Windows 上安装 Python 3 后运行:
```bash
python server.py --nano-host <Nano_IP> --http-port 8088 --udp-port 15001
```
浏览器打开:
```text
http://127.0.0.1:8088
```
## 5. UDP 命令格式
### 切换模式
```json
{"type":"mode","mode":"REMOTE"}
```
```json
{"type":"mode","mode":"WEB"}
```
```json
{"type":"mode","mode":"DISABLED"}
```
### Web 速度控制
```json
{
"type": "cmd_vel",
"linear": {"x": 0.2, "y": 0.0, "z": 0.0},
"angular": {"x": 0.0, "y": 0.0, "z": 0.1}
}
```
Nano 端会再次限幅:
```text
vx <= ±0.8 m/s
vy <= ±0.3 m/s
yaw <= ±0.5 rad/s
```
### 零速度
```json
{"type":"zero"}
```
### 软急停
```json
{"type":"estop","data":true}
```
## 6. 安全保护
当前最小版本已经包含:
1. 遥控器误触发阈值:`40`
2. 遥控器/Web/Nav 互斥控制模式。
3. `cmd_mux` 二次限幅。
4. `cmd_mux` 加速度限制。
5. Web UDP 超时自动发布零速度。
6. 急停优先级最高。
7. Web 页面切换到 `WEB` 模式需要确认。
8. Web 松开虚拟摇杆会自动发送零速度。
建议实机调试流程:
1. 先点击 `DISABLED`
2. 确认 `/cmd_vel` 为零。
3. 如果使用遥控器,点击 `REMOTE`
4. 如果使用 Web,点击 `WEB` 并确认周围安全。
5. 一旦异常,立即点击 `软急停`
## 7. 验证命令
查看最终输出速度:
```bash
ros2 topic echo /cmd_vel
```
查看遥控器输入:
```bash
ros2 topic echo /cmd_vel_remote
```
查看 Web 输入:
```bash
ros2 topic echo /cmd_vel_web
```
查看当前仲裁模式:
```bash
ros2 topic echo /control/mode_state
```
查看策略状态:
```bash
ros2 topic echo /runtime/target --field target_source
```
@@ -0,0 +1,20 @@
Tcl_0: [-0.009160, -0.999960, 0.000320, 0.032150,
0.002390, -0.000340, -1.000000, -0.011850,
0.999960, -0.009160, 0.002390, 0.005360,
0.000000, 0.000000, 0.000000, 1.000000]
cam_0:
image_width: 1600
image_height: 1296
k2: 0.000656
k3: -0.028961
k4: 0.045390
k5: -0.064513
k6: 0.038735
k7: -0.009903
p1: 0.000000
p2: 0.000000
A11: 736.894262
A12: -0.161150
A22: 736.611354
u0: 806.125535
v0: 639.650710
File diff suppressed because it is too large Load Diff
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,9 @@
# `v0.12.0` 里程计导航点云
本阶段使用三份 ASCII PCD 进行路线和里程计导航联调。原始文件均使用 `tools/nav_tools/downsample_ascii_pcd.py` 确定性等步长抽样,抽样文件仅用于可视化、路线编辑和软件回归,不等价于原始高密度地图。
| 文件 | 原始点数 | 步长 | 抽样点数 | 抽样大小 | 原始 SHA-256 | 抽样 SHA-256 |
| --- | ---: | ---: | ---: | ---: | --- | --- |
| `map_b.pcd` | 1,080,047 | 4 | 270,012 | 8,680,283 B | `F17FC7E188C772D48E8D3FE20E3A934AFCDF59357C27BCCDCAC3C7E9AD1D1CB7` | `F3052431F217DE8FFFE87726F8E547A59C65048B0285F02AF4803C5F4E4C1B42` |
| `A_min.pcd` | 683,908 | 3 | 227,969 | 7,300,173 B | `77C884B17D78C4D157F84093BFEA0729FA25382C70E706C4BA12913A3C90ED20` | `CC4556D472BE6E5B232B9BFEE43F5F337D710F88E88D9B90994368C580327DC1` |
| `C.pcd` | 3,186,653 | 11 | 289,696 | 9,040,786 B | `C5C7E60B06A0FFEA15E55C14B8CFAFBC8525371C8C3AC9C72821AB736AED8A6C` | `1FC4124642D934460811DBF62EBE9C5C9CF30668BC42E4782B18F3C002ED1A25` |
File diff suppressed because it is too large Load Diff
@@ -0,0 +1,276 @@
{
"name": "A_min_route",
"map": "A_min",
"frame_id": "map",
"createdAt": "2026-07-04T00:40:56",
"segments": [
{
"name": "segment_1",
"obstacle": "slalom",
"waypoints": [
{
"id": 1,
"x": 9.3549,
"y": 0.9079,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 2,
"x": 11.1136,
"y": 0.8621,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 3,
"x": 13.0,
"y": 0.75,
"yawDeg": 88.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 4,
"x": 12.9102,
"y": 2.611,
"yawDeg": 90.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 5,
"x": 12.9102,
"y": 4.88,
"yawDeg": 190.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 6,
"x": 11.94,
"y": 4.6203,
"yawDeg": 178.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 7,
"x": 10.65,
"y": 4.65,
"yawDeg": 178.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 8,
"x": 9.3997,
"y": 4.649,
"yawDeg": -95.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 9,
"x": 9.3092,
"y": 3.8864,
"yawDeg": 180.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 10,
"x": 7.9263,
"y": 4.0028,
"yawDeg": 150.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 11,
"x": 6.8148,
"y": 4.6619,
"yawDeg": 170.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 12,
"x": 5.0054,
"y": 5.0,
"yawDeg": 180.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 13,
"x": 3.2994,
"y": 5.04,
"yawDeg": 185.0,
"speed": 0.7,
"policy": "crawl",
"tolerance": 0.15
},
{
"id": 14,
"x": 1.8838,
"y": 4.8455,
"yawDeg": 190.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 15,
"x": 1.06,
"y": 4.45,
"yawDeg": 270.0,
"speed": 0.35,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 16,
"x": 1.0561,
"y": 3.9,
"yawDeg": -90.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 17,
"x": 1.04,
"y": 2.79,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 18,
"x": 1.552,
"y": 2.77,
"yawDeg": -90.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 19,
"x": 1.519,
"y": 1.6677,
"yawDeg": 180.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 20,
"x": 0.4708,
"y": 1.6677,
"yawDeg": -90.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 21,
"x": 0.4552,
"y": 0.5,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 22,
"x": 1.2783,
"y": 0.5,
"yawDeg": 90.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 23,
"x": 1.2628,
"y": 1.6211,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 24,
"x": 2.3576,
"y": 1.5978,
"yawDeg": -90.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 25,
"x": 2.3265,
"y": 0.5,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 26,
"x": 3.12,
"y": 0.5,
"yawDeg": 45.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 27,
"x": 3.8639,
"y": 0.9844,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 28,
"x": 6.8222,
"y": 0.9145,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
},
{
"id": 29,
"x": 9.5321,
"y": 0.9223,
"yawDeg": 0.0,
"speed": 0.7,
"policy": "rough",
"tolerance": 0.15
}
]
}
],
"yawToleranceDegDefault": 30.0
}
@@ -0,0 +1,12 @@
# `v0.12.0` 策略文件
本阶段运行配置使用 `model_9600` 作为 Rough 策略,并保留 Crawl ONNX 和部署契约参考模型。
| 文件 | 用途 |
| --- | --- |
| `model_9600.onnx` | 里程计版本 Rough ONNX |
| `model_9600_fp16.engine` | 里程计版本 Rough TensorRT |
| `model_crawl.onnx` | Crawl 候选;当前配置使用 IK 后端 |
| `model_rough.onnx` | 原部署契约参考基线 |
其余 `model_6800``model_8400``model_10200` 等候选策略属于相邻实验或后续比赛版本,不在本 Tag 重复归档。
@@ -0,0 +1,8 @@
v0.10.5 2026_0525
1. 修改 recorddata 数据格式,新增 device_id、algorithm_version 字段
2. 修复 SLAM 模式下下载地图失败的问题(USB2.0)
3. 修复重定位模式下上传地图失败的问题(USB2.0)
v0.10.4 2026_0522
1. 修复 USB2.0 心跳超时导致软断开的问题
2. control_command.yaml 新增 custom_init_pose_search_radius 和 custom_init_pose_max_rot_deg 参数
@@ -144,27 +144,12 @@ if(ROS_VERSION STREQUAL "ROS1")
cv_bridge
tf
image_transport
message_generation
)
# ---- AE/AWB debug services ----
# Declare custom .srv files; catkin generates ros1 message headers for them.
add_service_files(
FILES
GetAe.srv
GetAwb.srv
SetAe.srv
SetAwb.srv
)
generate_messages(
DEPENDENCIES
std_msgs
)
include_directories(${catkin_INCLUDE_DIRS})
catkin_package(
CATKIN_DEPENDS roscpp std_msgs sensor_msgs nav_msgs cv_bridge image_transport message_runtime
CATKIN_DEPENDS roscpp std_msgs sensor_msgs nav_msgs cv_bridge image_transport
INCLUDE_DIRS include
)
@@ -189,9 +174,6 @@ if(ROS_VERSION STREQUAL "ROS1")
pthread
usb-1.0
)
# Make sure the AE/AWB srv headers are generated before host_sdk_sample
# is compiled (catkin's generate_messages produces this target).
add_dependencies(host_sdk_sample ${PROJECT_NAME}_generate_messages_cpp)
add_library(pointcloud_depth_converter src/pointcloud_depth_converter.cpp)
target_link_libraries(pointcloud_depth_converter
@@ -267,18 +249,6 @@ elseif(ROS_VERSION STREQUAL "ROS2")
find_package(tf2_ros REQUIRED)
find_package(geometry_msgs REQUIRED)
find_package(tf2_geometry_msgs REQUIRED)
find_package(ament_index_cpp REQUIRED)
# ---- AE/AWB debug services ----
# Generate C++ bindings for the 4 .srv files in srv/. The generated
# headers are linked into host_sdk_sample so it can host the services.
find_package(rosidl_default_generators REQUIRED)
rosidl_generate_interfaces(${PROJECT_NAME}
"srv/GetAe.srv"
"srv/GetAwb.srv"
"srv/SetAe.srv"
"srv/SetAwb.srv"
)
# Create executable
add_executable(host_sdk_sample
@@ -295,12 +265,6 @@ elseif(ROS_VERSION STREQUAL "ROS2")
usb-1.0
)
# Link the generated AE/AWB srv typesupport into host_sdk_sample so it
# can host the 4 debug services.
rosidl_get_typesupport_target(cpp_typesupport_target
${PROJECT_NAME} "rosidl_typesupport_cpp")
target_link_libraries(host_sdk_sample "${cpp_typesupport_target}")
# Add ROS2 dependencies
ament_target_dependencies(host_sdk_sample
rclcpp
@@ -314,7 +278,6 @@ elseif(ROS_VERSION STREQUAL "ROS2")
tf2_ros
geometry_msgs
tf2_geometry_msgs
ament_index_cpp
)
add_library(pointcloud_depth_converter_ros2 src/pointcloud_depth_converter.cpp)
@@ -403,9 +366,6 @@ elseif(ROS_VERSION STREQUAL "ROS2")
pcd2depth_ros2_node
cloud_reprojection_ros2_node
image_overlay_node
depth_image_ros2_node_lib
pointcloud_depth_converter_ros2
cloud_reprojector_ros2
EXPORT export_${PROJECT_NAME}
ARCHIVE DESTINATION lib
LIBRARY DESTINATION lib
@@ -459,7 +419,6 @@ elseif(ROS_VERSION STREQUAL "ROS2")
tf2_ros
geometry_msgs
tf2_geometry_msgs
ament_index_cpp
)
ament_package()

Some files were not shown because too many files have changed in this diff Show More