ttimelord/rl_course_project

Learning to Kick for a Bipedal Robot with Deep Reinforcement Learning

Learning to Kick for a Bipedal Robot with Deep Reinforcement Learning

Project for Prof. Shengbo Eben Li's Reinforcement Learning and Control course in Tsinghua University.

Installation

1. clone this repo

2. install pytorch

   https://pytorch.org/get-started/locally/

3. install Stable-Baselines3

pip install stable-baselines3[extra]

(version 1.8.0) Documentation: https://stable-baselines3.readthedocs.io/

Tutorial: https://github.com/araffin/rl-tutorial-jnrr19/blob/sb3

4. install pybullet

pip install pybullet

Documentation: https://docs.google.com/document/d/10sXEhzFRSnvFcl3XxNGhnD4N2SedqwdAvK3dsihxVUA/

Code Structure

• rl_course_project
  • assets/
    store URDF file of the robot
  • envs/
    gym environments
    • kick_ball_env.py only task reward
    • kick_ball_stand_env.py task reward with fall termination
    • kick_ball_imitation_env task reward, termination and imitation reward
  • logs/
    tensorboard logs
  • models/
    saved models
  • test_urdf.py
    test simulation and urdf
  • train.py
    use stable-baselines3 to train RL agnet
  • test.py
    Test trained model
  • config.py
    parameters of the robot and simulation