EvoRL is an efficient Evolutionary Reinforcement Learning Framework based on EvoX (A JAX-accelerated EC Framework)
conda|mamba env create -f xuanwu.yaml
conda activate xuanwu
EvoRL uses hydra to manage configs and run algorithms.
Train agents from cli:
python -m evorl.train agent=a2c env=brax/ant
python -m evorl.train agent=a2c env=gymnax/CartPole-v1 agent_network.continuous_action=false
# sweep over multiple config values (seed=114 or seed=514):
python -m evorl.train -m agent=a2c env=brax/ant seed=114,514
- brax
- acme
- evox
- gymnax
- jumanji
- jaxmarl