Yes, Q-learning Helps Offline In-Context RL
This work addresses the problem of improving offline in-context reinforcement learning for researchers and practitioners by showing that RL objectives enhance performance, though it is incremental as it builds on existing methods.
The study tackled the limitations of supervised training objectives in offline in-context reinforcement learning by integrating RL objectives, resulting in an average performance improvement of about 30% over Algorithm Distillation across various datasets and a doubling of performance in a challenging environment.
Existing offline in-context reinforcement learning (ICRL) methods have predominantly relied on supervised training objectives, which are known to have limitations in offline RL settings. In this study, we explore the integration of RL objectives within an offline ICRL framework. Through experiments on more than 150 GridWorld and MuJoCo environment-derived datasets, we demonstrate that optimizing RL objectives directly improves performance by approximately 30% on average compared to widely adopted Algorithm Distillation (AD), across various dataset coverages, structures, expertise levels, and environmental complexities. Furthermore, in the challenging XLand-MiniGrid environment, RL objectives doubled the performance of AD. Our results also reveal that the addition of conservatism during value learning brings additional improvements in almost all settings tested. Our findings emphasize the importance of aligning ICRL learning objectives with the RL reward-maximization goal, and demonstrate that offline RL is a promising direction for advancing ICRL.