RoboHive is a collection of environments/tasks simulated with the MuJoCo physics engine exposed using the OpenAI-Gym API. Its compatible with any gym-compatible agents training framework (Stable Baselines, RLlib, TorchRL, AgentHive, etc)

Getting started with RoboHive is as simple as -

# Install RoboHive
pip install robohive
# Initialize RoboHive
robohive_init
# Demo an environment
python -m robohive.utils.examine_env -e FrankaReachRandom-v0

or, alternatively for editable installation -

# Clone RoboHive
git clone --recursive https://github.com/vikashplus/robohive.git; cd robohive
# Install (editable) RoboHive
pip install -e .
# Demo an environment
python -m robohive.utils.examine_env -e FrankaReachRandom-v0

See detailed installation instructions for options on mujoco-python-bindings and visual-encoders (R3M, RRL, VC), and frequently asked questions for more details.

RoboHive contains a variety of environments, which are organized as suites. Each suite is a collection of loosely related environments. The following suites are provided at the moment with plans to improve the diversity of the collection.

Hand-Manipulation-Suite (video)
A collection of environments centered around dexterous manipulation. Standard ADROIT benchmarks introduced in Learning Complex Dexterous Manipulation with Deep Reinforcement Learning and Demonstrations, RSS2018.) are a part of this suite

Arm-Manipulation-Suite
A collection of environments centered around Arm manipulation.

Myo-Suite (website)
A collection of environments centered around Musculoskeletal control.

Myo/MyoDM-Suite (Website)
A collection of musculoskeletal environments for dexterous manipulation introduced as MyoDM in MyoDeX.

MultiTask Suite
A collection of environments centered around multi-task. Standard RelayKitchen benchmarks are a part of this suite.

This suite contains a collection of environments centered around dexterous manipulation. Standard TCDM benchmarks are a part of this suite

This suite contains a collection of environments centered around real-world locomotion and manipulation. Standard ROBEL benchmarks are a part of this suite

If you find RoboHive useful in your research,

• please consider supporting the project by providing a star ⭐
• please consider citing our project by using the following BibTeX entry:

@Misc{RoboHive2020,
  title = {RoboHive -- A Unified Framework for Robot Learning},
  howpublished = {\url{https://sites.google.com/view/robohive}},
  year = {2020},
  url = {https://sites.google.com/view/robohive},
}