This repository contains an environment for simulating a quadruped robot.
Recommend using a virtualenv (or conda) with python3.6 or higher. After installing virtualenv with pip, this can be done as follows:
virtualenv {quad_env, or choose another name venv_name} --python=python3
To activate the virtualenv:
source {PATH_TO_VENV}/bin/activate
Your command prompt should now look like:
(venv_name) user@pc:path$
The repository depends on recent versions of pybullet, gym, numpy, stable-baselines3, matplotlib, etc., pip install [PACKAGE]
- env for the quadruped environment files, please see the gym simulation environment quadruped_gym_env.py, the robot specific functionalities in quadruped.py, and config variables in configs_a1.py. You will need to make edits in quadruped_gym_env.py, and review quadruped.py carefully for accessing robot states and calling functions to solve inverse kinematics, return the leg Jacobian, etc.
- a1_description contains the robot mesh files and urdf.
- utils for some file i/o and plotting helpers.
- hopf_network.py provides a CPG class skeleton for various gaits, and run_cpg.py maps these joint commands to be executed on an instance of the quadruped_gym_env class. Please fill in these files carefully.
- run_sb3.py and load_sb3.py provide an interface to training RL algorithms based on stable-baselines3. You should review the documentation carefully for information on the different algorithms and training hyperparameters.
- The PyBullet Quickstart Guide is the current up-to-date documentation for interfacing with the simulation.
- The quadruped environment took inspiration from Google's motion-imitation repository based on this paper.
- Reinforcement learning algorithms from stable-baselines3. Also see for example ray[rllib] and spinningup.
The CPG and RL framework are based on the following papers:
- G. Bellegarda and A. Ijspeert, "CPG-RL: Learning Central Pattern Generators for Quadruped Locomotion," in IEEE Robotics and Automation Letters, 2022, doi: 10.1109/LRA.2022.3218167. IEEE, arxiv
- G. Bellegarda, Y. Chen, Z. Liu, and Q. Nguyen, "Robust High-speed Running for Quadruped Robots via Deep Reinforcement Learning," in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2022. arxiv
- If your simulation is very slow, remove the calls to time.sleep() and disable the camera resets in quadruped_gym_env.py.
- The camera viewer can be modified in
_render_step_helper()in quadruped_gym_env.py to track the hopper.