Robogym is a robotics simulation framework designed for robot learning research, offering a variety of environments built on OpenAI Gym and MuJoCo. It targets researchers and engineers needing diverse, customizable simulation scenarios for training and evaluating reinforcement learning agents, particularly for manipulation tasks.

How It Works

Robogym leverages MuJoCo for physics simulation and extends OpenAI Gym's interface, providing a structured way to define and interact with robotic environments. Environments are built using a SimulationInterface and RobotEnv classes, allowing for detailed customization of robot configurations, physics parameters, and task objectives through RobotEnvParameters and RobotEnvConstants. This approach facilitates domain randomization and curriculum learning by enabling direct intervention in environment parameters during training.

Quick Start & Requirements

• Install MuJoCo following mujoco-py instructions.
• Install Robogym: pip install git+https://github.com/openai/robogym.git
• Requires Python 3.7.4+. Tested on macOS Mojave/Catalina and Ubuntu 16.04 LTS.
• Visualize environments: python robogym/scripts/examine.py <env_script>
• Official documentation: Not explicitly linked, but usage examples are provided.

Highlighted Details

• Includes "Dactyl" environments simulating a Shadow Robot hand for in-hand manipulation.
• Features "Rearrange" environments with a UR16e robot for tabletop object manipulation tasks.
• Supports customization of environment parameters (e.g., randomize, mujoco_substeps, num_objects) for domain randomization and curriculum learning.
• Offers teleoperation capabilities for interactive environment testing and custom environment creation.

Maintenance & Community

• The project is marked as "Archive" and no further updates are expected.
• Developed by OpenAI.

Licensing & Compatibility

• The repository does not explicitly state a license. Based on OpenAI's typical practices, it's likely intended for research use. Compatibility with commercial or closed-source projects is not specified.

Limitations & Caveats

This project is archived, meaning it will not receive updates or bug fixes. The dependency on MuJoCo and specific Python versions may lead to compatibility issues with newer systems.