Mildly Conservative Q-Learning for Offline Reinforcement Learning

TL;DR

This paper introduces Mildly Conservative Q-learning (MCQ), a novel offline RL method that balances conservatism and generalization, leading to improved performance and transferability on benchmark tasks.

Contribution

MCQ actively trains OOD actions with pseudo Q values, providing a theoretically justified approach that enhances offline RL performance without excessive pessimism.

Findings

MCQ outperforms prior methods on D4RL benchmarks.

MCQ demonstrates superior transfer from offline to online settings.

MCQ maintains conservative estimates without overestimating OOD actions.

Abstract

Offline reinforcement learning (RL) defines the task of learning from a static logged dataset without continually interacting with the environment. The distribution shift between the learned policy and the behavior policy makes it necessary for the value function to stay conservative such that out-of-distribution (OOD) actions will not be severely overestimated. However, existing approaches, penalizing the unseen actions or regularizing with the behavior policy, are too pessimistic, which suppresses the generalization of the value function and hinders the performance improvement. This paper explores mild but enough conservatism for offline learning while not harming generalization. We propose Mildly Conservative Q-learning (MCQ), where OOD actions are actively trained by assigning them proper pseudo Q values. We theoretically show that MCQ induces a policy that behaves at least as well as the behavior policy and no erroneous overestimation will occur for OOD actions. Experimental results on the D4RL benchmarks demonstrate that MCQ achieves remarkable performance compared with prior work. Furthermore, MCQ shows superior generalization ability when transferring from offline to online, and significantly outperforms baselines. Our code is publicly available at https://github.com/dmksjfl/MCQ.