Dopamine
An open-source reinforcement learning research framework with JAX agents, Atari and MuJoCo environments, and Docker support for local experiments.
Dopamine is a reinforcement learning framework for researchers who want to modify agents and test algorithm ideas. Its compact codebase focuses on a limited set of established algorithms and supports experiments on your own hardware.
- Editable agents: Work with JAX implementations of DQN, C51, Rainbow, IQN, SAC and PPO, and change the code directly for your experiments.
- Atari and MuJoCo: Train and evaluate agents in these supported environments for reinforcement learning research.
- Reference baselines: Compare experimental results with supplied baselines for each agent, using an experiment setup designed around published Arcade Learning Environment evaluation recommendations.
JAX is the actively maintained backend. Many agents also have legacy TensorFlow implementations, so backend availability depends on the algorithm. Colaboratory notebooks provide examples alongside the locally runnable framework, and Docker containers are available.
Dopamine is open source under Apache 2.0. It's designed for research that involves changing the implementation itself, rather than using a fixed application interface.
Logotyper
