Code for "Efficient and Stable Off-policy Training via Behavior-aware Evolutionary Learning" accepted by CoRL2022.
It is straight forward to run BEL in a local conda environment. First make sure you have mujoco200 and mjkey.txt in your ~/.mujoco. Then create the conda env and run bel.py. By default, experiment's logs will be output to a directory called results.
conda env create -f environment.yml
conda activate bel
python bel.py --env_name Ant-v3 --seed 10