Code implementation of the benchmarks from the following report.
This benchmark compares the old version of Gym-Robotics environments that depend on mujoco_py
and the latest version of the environments
that use the lates official python bindings from mujoco.
The benchmark uses an adapted version of TQC + HER from Stable Baselines 3 and the hyperparameters given in rl-baselines3-zoo
To run the benchmark first intall the dependencies:
pip install -r requirements.txt
Then to reproduce the examples from the FetchReach
environment run:
OMP_NUM_THREADS=1 xvfb-run -a python main.py \
--env-id FetchReach-v1 FetchReach-v2 \
--command "python sb3_tqc_her.py" \
--num-seeds 5 \
--workers 10