Survival-based multi-agent reinforcement learning exploration.
The environments built in this repository depend on the MuJoCo physics engine. Once a license is obtained and requirements are installed, run the application to train an model on an environment, or view an environment, e.g.:
python3 darwin/runner.py darwin/envs/simple_food.py --policy-name=dqn
Live demos are available on darwin-app.web.app.
Environment building is inpsired by the environment design done by OpenAI for their paper Emergent Tool Use From Multi-Agent Autocurricula.