RAD: Retrieval High-quality Demonstrations to Enhance Decision-making
Updated
Updated · arxiv.org · Jul 17
RAD: Retrieval High-quality Demonstrations to Enhance Decision-making
1 articles · Updated · arxiv.org · Jul 17
Summary
Researchers have introduced RAD, a new retrieval-augmented method to enhance offline reinforcement learning policy generalization.
RAD retrieves high-return, reachable states from static datasets and uses generative models to plan sub-trajectories, outperforming existing baselines in diverse benchmarks.
This approach addresses the limitations of static data in offline RL, potentially improving decision-making in fields like robotics, healthcare, and autonomous driving.