Inverse reinforcement learning for dexterous hand manipulation

This code accompanies the paper "Inverse reinforcement learning for dexterous hand manipulation" to enable reproduction of the presented results.

Installation steps (Linux or MacOS):

For mujoco-py:

• Install mujoco libraries: https://www.roboti.us/index.html
• sudo apt install libglew-dev patchelf libosmesa6-dev

If running on Linux

Add to .zshrc:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/<user>/.mujoco/mujoco200/bin
export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLEW.so

If running on MacOS

Need to install gcc for mujoco-py 2.0:
brew install gcc

Install package

pip3 install -e . && pip3 install -e git+https://github.com/Jendker/mj_envs.git#egg=mj_envs

Results reproduction

Training can be started with:
cd inverse_rl_dexterous_hand/training
python training.py --configs IRL

Please see the 'inverse_rl_dexterous_hand/training/README.md' for the detailed commands for results reproduction.