D4RL

D4RL is a benchmark that contains a variety of offline reinforcement learning tasks used to evaluate the performance of algorithms in generating policies and learning from previously collected data.

Papers using D4RL (3)

One-Step Generative Policies with Q-Learning: A Reformulation of MeanFlow2025Task-agnostic Decision Transformer for Multi-type Agent Control with Federated Split Training2024Planning Transformer: Long-Horizon Offline Reinforcement Learning with Planning Tokens2024