DYROS TOCABI RL Locomotion Repository

This repository is dedicated to DYROS TOCABI RL Locomotion. It has been developed based on the NVIDIA IsaacGym's Preview4, which can be accessed here: NVIDIA IsaacGym.

For environment setup, execute create_conda_env_rlgpu.sh. This script creates a conda environment named rlgpu4. After activating the rlgpu4 environment, run the following command in the python directory:

pip install -e .

Then, in the python/IsaacGymEnvs directory, execute the following:

pip install -e .

Training Your First TOCABI Policy

To train your first TOCABI policy, execute the following command in the python/IsaacGymEnvs/isaacgymenvs directory:

python train.py task=DyrosDynamicWalk

Inference with a Trained Model

To load a trained checkpoint and only perform inference (no training), pass test=True as an argument, along with the checkpoint name. To avoid rendering overhead, you may also want to run with fewer environments using num_envs=64:

python train.py task=DyrosDynamicWalk checkpoint=runs/DyrosDynamicWalk/nn/DyrosDynamicWalk.pth test=True num_envs=64 headless=False

Expected Error

1. ImportError: libpython3.8.so.1.0: cannot open shared object file: No such file or directory
   • Solution:

   export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/$USER_NAME/anaconda3/envs/rlgpu4/lib

General Training Recipes for DYROS

Refer to General Recipe for DYROS