TROT-Q

ASCC2022 - Eungchang Mason Lee, Jinwoo Jeon, and Hyun Myung "TROT-Q: TRaversability and Obstacle aware Target tracking system for Quadruped robots"

• Tracking detected target (YOLOv4-tiny-3l network is used)
• FAST-LIO2 SLAM is used to estimate the states
• Traversability and obstacle collision are checked
• Minimum Jerk Trajectory is tracked
• High-level velocity MPC control
• Low-level PID control

Video - https://youtu.be/hEQSOyfWevY

Dependencies

• OpenCV version >= 4.4.0
• cv_bridge
• ROS and Gazebo
  • refer here
  • $ sudo apt install ros-<distro>-desktop-full
• Dependencies

$ sudo apt install ros-melodic-gazebo-plugins ros-melodic-ros-control ros-melodic-ros-controllers
$ sudo apt install ros-melodic-hector-gazebo-plugins

$ wget -O ubuntu.sh https://raw.githubusercontent.com/PX4/PX4-Autopilot/master/Tools/setup/ubuntu.sh
$ wget https://raw.githubusercontent.com/PX4/PX4-Autopilot/master/Tools/setup/requirements.txt
$ source ubuntu.sh
$ sudo apt upgrade libignition-math4

• This Repo

$ cd ~/your_workspace/src

$ git clone --recursive https://github.com/engcang/trot-q

$ cd trot-q/gazebo_maps
$ tar -xf height_maze.tar.xz
$ tar -xf common_models.tar.xz

$ cd ..
$ echo "export GAZEBO_MODEL_PATH=$GAZEBO_MODEL_PATH:$(pwd)/gazebo_maps/height_maze:$(pwd)/gazebo_maps/common_models" >> ~/.bashrc
$ . ~/.bashrc

$ cd ~/your_workspace
$ catkin build -DCMAKE_BUILD_TYPE=Release

How to run

• Launch Gazebo world
  • Change world in the launch file, if you want.
  • Check dev of joystick to manually control Jackal in the launch file.

$ roslaunch trot-q gazebo.launch

• Run TROT-Q

$ roslaunch trot-q main.launch

Components

• ANYMAL-B Gazebo model from here
  • Many unnecessary files are trimmed
• Champ quadruped robot controller from here
• Gazebo map, made by myself from here
• Gazebo Jackal Gazebo model from here
  • Many unnecessary files are trimmed
• Gazebo Ouster - referred here1 and here2
  • Many unnecessary files are trimmed
  • GPU-ray is enabled as here

SLAM

• FAST-LIO2

Object Detection

• ROS-YOLO using OpenCV code: from here (myself)