ACL-QL: Adaptive Conservative Level in Q-Learning for Offline Reinforcement Learning

TL;DR

ACL-QL introduces an adaptive conservative control mechanism in offline reinforcement learning, allowing nuanced Q-value regulation per state-action pair, leading to improved performance and safety over fixed conservative methods.

Contribution

The paper presents a novel framework and algorithm for adaptively controlling conservative levels in Q-learning, addressing over-conservatism and fixed constraint issues in offline RL.

Findings

Achieves state-of-the-art results on D4RL benchmarks.

Effectively balances conservatism and performance.

Demonstrates the benefits of adaptive conservative control.

Abstract

Offline Reinforcement Learning (RL), which operates solely on static datasets without further interactions with the environment, provides an appealing alternative to learning a safe and promising control policy. The prevailing methods typically learn a conservative policy to mitigate the problem of Q-value overestimation, but it is prone to overdo it, leading to an overly conservative policy. Moreover, they optimize all samples equally with fixed constraints, lacking the nuanced ability to control conservative levels in a fine-grained manner. Consequently, this limitation results in a performance decline. To address the above two challenges in a united way, we propose a framework, Adaptive Conservative Level in Q-Learning (ACL-QL), which limits the Q-values in a mild range and enables adaptive control on the conservative level over each state-action pair, i.e., lifting the Q-values more for good transitions and less for bad transitions. We theoretically analyze the conditions under which the conservative level of the learned Q-function can be limited in a mild range and how to optimize each transition adaptively. Motivated by the theoretical analysis, we propose a novel algorithm, ACL-QL, which uses two learnable adaptive weight functions to control the conservative level over each transition. Subsequently, we design a monotonicity loss and surrogate losses to train the adaptive weight functions, Q-function, and policy network alternatively. We evaluate ACL-QL on the commonly used D4RL benchmark and conduct extensive ablation studies to illustrate the effectiveness and state-of-the-art performance compared to existing offline DRL baselines.