Action Candidate Based Clipped Double Q-learning for Discrete and Continuous Action Tasks

TL;DR

This paper introduces an action candidate based clipped double estimator to improve the accuracy of maximum expected action value estimation in Double Q-learning, reducing bias and enhancing performance in both discrete and continuous tasks.

Contribution

It proposes a novel action candidate based estimator that reduces underestimation bias and extends to continuous actions, improving over traditional clipped Double Q-learning.

Findings

More accurate maximum expected action value estimation in toy environments.

Better performance on benchmark problems.

Bias control via the number of action candidates.

Abstract

Double Q-learning is a popular reinforcement learning algorithm in Markov decision process (MDP) problems. Clipped Double Q-learning, as an effective variant of Double Q-learning, employs the clipped double estimator to approximate the maximum expected action value. Due to the underestimation bias of the clipped double estimator, performance of clipped Double Q-learning may be degraded in some stochastic environments. In this paper, in order to reduce the underestimation bias, we propose an action candidate based clipped double estimator for Double Q-learning. Specifically, we first select a set of elite action candidates with the high action values from one set of estimators. Then, among these candidates, we choose the highest valued action from the other set of estimators. Finally, we use the maximum value in the second set of estimators to clip the action value of the chosen action in the first set of estimators and the clipped value is used for approximating the maximum expected action value. Theoretically, the underestimation bias in our clipped Double Q-learning decays monotonically as the number of the action candidates decreases. Moreover, the number of action candidates controls the trade-off between the overestimation and underestimation biases. In addition, we also extend our clipped Double Q-learning to continuous action tasks via approximating the elite continuous action candidates. We empirically verify that our algorithm can more accurately estimate the maximum expected action value on some toy environments and yield good performance on several benchmark problems.