Collection of MuJoCo robotics environments equipped with both vision and tactile sensing.
The environments are built on top of Gymnasium.

Current available environments:

• tactile_envs/Insertion-v0 (see Insertion)
• Door (see Door)
• HandManipulateBlockRotateZFixed-v1 (see HandManipulateBlockRotateZFixed)
• HandManipulateEggRotateFixed-v1 (see HandManipulateEggRotateFixed)
• HandManipulatePenRotateFixed-v1 (see HandManipulatePenRotateFixed)

To install tactile_envs in a fresh conda env:

conda create --name tactile_envs python=3.11
conda activate tactile_envs
pip install -r requirements.txt

Before running the environment code, make sure that you generate the tactile sensor collision meshes for the desired resolution. E.g., for 32x32 sensors:

python tactile_envs/assets/insertion/generate_pad_collisions.py --nx 32 --ny 32

python scripts/test_env_insertion.py

If you use these environments in your research, please cite the following paper:

@article{sferrazza2023power,
  title={The power of the senses: Generalizable manipulation from vision and touch through masked multimodal learning},
  author={Sferrazza, Carmelo and Seo, Younggyo and Liu, Hao and Lee, Youngwoon and Abbeel, Pieter},
  year={2023}
}

Are you interested in more complex robot environments with high-dimensional tactile sensing? Check out HumanoidBench!

This codebase contains some files adapted from other sources:

• Gymnasium-Robotics: https://github.com/Farama-Foundation/Gymnasium-Robotics
• robosuite: https://github.com/ARISE-Initiative/robosuite/tree/master
• TactileSimulation: https://github.com/eanswer/TactileSimulation