All Tools
D
AgentsFreeOpen Source
D4RL
Offline RL datasets and reference environments
Apache-2.0
ABOUT
Offline RL needs fixed datasets, but labs used to collect their own rollouts and could not compare algorithms. D4RL attaches a standard dataset to each Gym task via env.get_dataset so CQL, IQL, and similar methods train on the same observations and actions.
INSTALL
pip install d4rlINTEGRATION GUIDE
1. Train an offline RL algorithm on D4RL MuJoCo locomotion datasets
2. Compare CQL or IQL against published D4RL scores without new environment interaction
3. Load observations, actions, and terminals from env.get_dataset for custom learners
4. Evaluate offline policies on Maze2D, AntMaze, or Adroit manipulation tasks
TAGS
pythonoffline-rldatasetsgymmujocofarama