This repository provides open-source implementations of OpenAI Gym's MuJoCo environments using PyBullet, targeting reinforcement learning researchers. It aims to remove the licensing barrier of MuJoCo, enabling free and open research with continuous control tasks.

How It Works

The project reimplements popular MuJoCo environments using PyBullet's Python wrapper. This approach allows seamless integration with the OpenAI Gym framework, maintaining a consistent API for RL agents. The use of PyBullet, a free and open-source physics engine, eliminates the need for a commercial MuJoCo license, democratizing access to these challenging simulation environments.

Quick Start & Requirements

• Install OpenAI Gym: pip install -e . from cloned repo.
• Install PyBullet Gym: pip install -e . from cloned repo.
• Requires Python 2.7 or 3.5+, pip version 1.5.0+.
• Test: import gym; import pybulletgym; env = gym.make('HumanoidPyBulletEnv-v0'); env.render(); env.reset()
• Official Docs: https://github.com/benelot/pybullet-gym

Highlighted Details

• Provides PyBullet-based alternatives for many OpenAI Gym MuJoCo environments.
• Includes pretrained agents for several MuJoCo environments as baselines.
• Environments are structured into robot-scene combinations for flexibility.
• Roadmap includes porting OpenAI Robotics Gyms and DeepMind Control Suite.

Maintenance & Community

• Project initiated in 2018 by Benjamin Ellenberger.
• Roadmap indicates ongoing development and planned ports.

Licensing & Compatibility

• License: Not explicitly stated in the README, but implies open-source use.
• Compatibility: Designed for OpenAI Gym, supports Linux, Windows, OS X.

Limitations & Caveats

Some manipulator environments (striker, pusher, thrower) are marked as "WIP" with loading issues. Porting of some MuJoCo environments may have observation differences or missing features.