Intro

This is a simple 4-leg robot simulation model and integrated with reinforcement learning algorithm. Each of its leg has 4 DOF. The simulation platform is Gazebo using ROS control and openai_ros to integrate the model with the RL framework Gym.

Environment Setup

• ubuntu 20.04. To set up the system with only python3 sudo apt install python-is-python3
• ROS-Noetic (Installation guide)
  • Gazebo (contained in the ros-noetic-desktop-full in the above guide)
  • effort controller sudo apt install ros-noetic-effort-controller
  • velocity controller sudo apt install ros-noetic-velocity-controller
  • position controller sudo apt install ros-noetic-position-controller
• Gym version 0.17.2 pip3 install gym
• Torch version 1.7.1 pip3 install torch

Remember follow this guide to create a ROS workspace. e.g. ~/catkin_ws

Clone the repo and move it to ~/catkin_ws/src. Then run catkin_make at ~/catkin_ws

Usage

Firstly, run the simulation, cmd in [] is optional
roslaunch crawler world.launch [gui:=false]

Then run the training script in the script dir.
python crawlerD.py