Decorrelated Double Q-learning
This addresses performance issues in reinforcement learning for continuous control, but it is incremental as it builds on Double Q-learning.
The paper tackles the problem of overestimation bias and imprecise estimates in Q-learning with value function approximation by introducing decorrelated double Q-learning (D2Q), which uses a decorrelated regularization item to reduce correlation between approximators, resulting in improved performance on MuJoCo continuous control tasks.
Q-learning with value function approximation may have the poor performance because of overestimation bias and imprecise estimate. Specifically, overestimation bias is from the maximum operator over noise estimate, which is exaggerated using the estimate of a subsequent state. Inspired by the recent advance of deep reinforcement learning and Double Q-learning, we introduce the decorrelated double Q-learning (D2Q). Specifically, we introduce the decorrelated regularization item to reduce the correlation between value function approximators, which can lead to less biased estimation and low variance. The experimental results on a suite of MuJoCo continuous control tasks demonstrate that our decorrelated double Q-learning can effectively improve the performance.