# dogtracker

**Repository Path**: tryhardjackma/dogtracker

## Basic Information

- **Project Name**: dogtracker
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-06-11
- **Last Updated**: 2026-06-11

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# GO2 RGB-D Person Following

This project variant keeps only the pieces needed to deploy a vision-driven
quadruped following algorithm:

- Unitree GO2 Gazebo/CHAMP locomotion stack
- RGB-D camera sensor on the robot model
- Simulated people in the Gazebo environment
- RT-DETR person detection
- LiteTrack-style lightweight target association
- Direct RGB-D target-to-`/cmd_vel` control

LiDAR, SLAM Toolbox, Nav2 planning, costmaps, scan relay, and map-based
navigation are no longer part of the default launch path.

## Runtime Flow

```mermaid
graph TD
    CAM[GO2 RGB-D camera]
    DET[RT-DETR person detector]
    TRK[LiteTrack target association]
    CTRL[RGB-D person_follower_node]
    CMD[/cmd_vel]
    CHAMP[CHAMP quadruped controller]
    GO2[GO2 motion]

    CAM --> DET
    DET --> TRK
    TRK -->|/detector/target| CTRL
    CTRL --> CMD
    CMD --> CHAMP
    CHAMP --> GO2
```

The controller uses the target position in `camera_color_optical_frame`:

- `position.z`: forward depth to maintain `desired_distance_m`
- `position.x`: lateral offset to turn the robot toward the person

## Key Files

```text
go2_description/xacro/go2_robot.xacro             GO2 model entry
go2_description/xacro/camera.xacro                RGB-D camera sensor
go2_description/xacro/accessories.urdf.xacro      IMU only; LiDAR removed from robot
go2_yolo_detector/go2_yolo_detector/rtdetr_litetrack_node.py
go2_yolo_navigator/go2_yolo_navigator/person_follower_node.py
go2_yolo_bringup/launch/gazebo_launch.py          Gazebo + GO2 + CHAMP
go2_yolo_bringup/launch/yolo_nav_launch.py        RT-DETR + LiteTrack + /cmd_vel
go2_yolo_bringup/launch/advanced_demo_launch.py   Simulated moving people
go2_yolo_bringup/config/detector_params.yaml      Perception/control parameters
```

Old training, evaluation, Nav2, SLAM, LiDAR, and YOLO files were moved under
`legacy_unused/` so they do not participate in the Humble build.

## Dependencies

Clone CHAMP next to this package in your ROS 2 workspace:

```bash
git clone https://github.com/chvmp/champ.git
git clone https://github.com/chvmp/champ_teleop.git
```

Install ROS and Python dependencies:

```bash
sudo apt install \
  ros-humble-robot-localization \
  ros-humble-ros2-control \
  ros-humble-ros2-controllers \
  ros-humble-gazebo-ros2-control \
  ros-humble-gazebo-ros-pkgs \
  ros-humble-xacro \
  ros-humble-joint-state-publisher \
  ros-humble-tf2-ros

pip install -r requirements.txt
```

## Build

```bash
mkdir -p ~/go2_follow_ws/src
cd ~/go2_follow_ws/src
git clone <this-repo-url>
git clone https://github.com/chvmp/champ.git
git clone https://github.com/chvmp/champ_teleop.git

cd ~/go2_follow_ws
rosdep install --from-paths src --ignore-src -r -y
colcon build --symlink-install
source install/setup.bash
```

## Run Simulation

Terminal 1: start Gazebo, GO2, RGB-D camera, and CHAMP control.

```bash
source install/setup.bash
ros2 launch go2_yolo_bringup gazebo_launch.py
```

Terminal 2: start RT-DETR + LiteTrack and direct following control.

```bash
source install/setup.bash
ros2 launch go2_yolo_bringup yolo_nav_launch.py \
  model_path:=rtdetr-l.pt \
  target_class:=person \
  desired_distance_m:=1.2
```

For the moving-person environment, launch Gazebo with the advanced world and then
start the people manager. Detection still comes from the RGB-D camera through
RT-DETR/LiteTrack:

```bash
ros2 launch go2_yolo_bringup gazebo_launch.py \
  world_file:=$(ros2 pkg prefix go2_yolo_bringup)/share/go2_yolo_bringup/worlds/advanced_demo_world.world

ros2 launch go2_yolo_bringup advanced_demo_launch.py
```

## Run On Robot

The real robot launch assumes the GO2 locomotion stack already consumes
`/cmd_vel`, and that the RGB-D camera publishes RealSense-style topics.

```bash
source install/setup.bash
ros2 launch go2_yolo_bringup real_robot_launch.py \
  model_path:=/path/to/rtdetr_person.pt \
  target_class:=person \
  desired_distance_m:=1.2
```

Adjust the camera remappings in
`go2_yolo_bringup/launch/real_robot_launch.py` if your camera driver uses
different topic names.

## Useful Topics

```bash
ros2 topic echo /detector/target
ros2 topic echo /follower/status
ros2 topic echo /cmd_vel
ros2 topic echo /detector/visualization
```

## Control Parameters

Important values in `go2_yolo_bringup/config/detector_params.yaml`:

- `model_path`: RT-DETR weights
- `target_class`: usually `person`
- `confidence_threshold`: minimum detection confidence
- `desired_distance_m`: stand-off distance from the person
- `linear_kp`, `angular_kp`: proportional control gains
- `max_linear_mps`, `max_angular_radps`: GO2 command limits
- `target_timeout_s`: stop if no recent tracked target is available

## Notes

- `go2_description/xacro/accessories.urdf.xacro` no longer instantiates the
  Hokuyo LiDAR.
- `legacy_unused/` is a non-destructive archive of files removed from the active
  build path.
- The retained `go2_yolo_msgs` messages still provide the common target
  interface between perception and control.