/deep-mimic-pybullet

Primary LanguagePython

Deep Mimic with PyBullet Environment

This repository contains a fork of PyBullet's implementation of the humanoid model used in DeepMimic, and our implementation of the RL pipeline to mimic motion from a reference motion clip on the humanoid model.

Install

Install PyTorch using conda or pip.

conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia

Then install the package.

pip install --upgrade pip
pip install -e .

Please check for an installation script in scripts/install.sh.

Run on Euler

Modules required to be loaded on Euler before running are:

module load gcc/8.2.0 python/3.9.9 cmake/3.25.0 freeglut/3.0.0 libxrandr/1.5.0  libxinerama/1.1.3 libxi/1.7.6  libxcursor/1.1.14 mesa/17.2.3 openmpi/4.1.4 eth_proxy

All sbatch scripts are in jobs/. Please run job scripts from the root directory of this repository since some paths coded in the scripts are relative.

Export

Videos can be exported from a trained model using:

python -m model.video_exporter -c <config>

Sample config files can be found in conf/.