RoboHive is a unified framework for robot learning, offering a collection of MuJoCo-simulated environments compatible with the OpenAI-Gym API. It targets researchers and engineers in reinforcement learning for robotics, providing a standardized interface for training agents across diverse manipulation and control tasks.

How It Works

RoboHive leverages the MuJoCo physics engine for realistic simulation and exposes environments through the OpenAI-Gym API, ensuring broad compatibility with popular RL agent training frameworks. This approach simplifies the integration of new RL algorithms and facilitates benchmarking by providing a consistent environment interface.

Quick Start & Requirements

• Install: pip install robohive
• Initialize: robohive_init
• Demo: python -m robohive.utils.examine_env -e FrankaReachRandom-v0
• Editable install: git clone --recursive https://github.com/vikashplus/robohive.git; cd robohive; pip install -e .
• Prerequisites: MuJoCo, Python. Optional: mujoco-python-bindings, visual-encoders (R3M, RRL, VC).
• Documentation: Detailed installation instructions

Highlighted Details

• Includes diverse suites: Hand-Manipulation, Arm-Manipulation, Myo-Suite, Myo/MyoDM-Suite, MultiTask Suite, TCDM Suite (WIP), ROBEL Suite (Coming soon).
• Supports standard benchmarks like ADROIT and RelayKitchen.
• Compatible with major RL frameworks: Stable Baselines, RLlib, TorchRL, AgentHive.

Maintenance & Community

• Project initiated in 2020.
• Community support via GitHub stars is encouraged.
• Citation details provided for academic use.

Licensing & Compatibility

• License not explicitly stated in the README.
• Compatibility with commercial or closed-source projects is not specified.

Limitations & Caveats

The TCDM Suite is marked as Work-In-Progress (WIP), and the ROBEL Suite is listed as "Coming soon," indicating ongoing development and potential for incomplete features or API changes in these areas.