Control Regularization for Reduced Variance Reinforcement Learning

TL;DR

This paper introduces a functional regularization method for model-free reinforcement learning that reduces variance, improves stability, and enhances learning efficiency in continuous control tasks.

Contribution

It proposes a novel functional regularization approach that regularizes policies in function space, with an adaptive strategy to optimize the bias-variance trade-off and preserve stability.

Findings

Significantly reduced variance in policy performance.

Enhanced dynamic stability during learning.

More efficient learning compared to standard deep RL.

Abstract

Dealing with high variance is a significant challenge in model-free reinforcement learning (RL). Existing methods are unreliable, exhibiting high variance in performance from run to run using different initializations/seeds. Focusing on problems arising in continuous control, we propose a functional regularization approach to augmenting model-free RL. In particular, we regularize the behavior of the deep policy to be similar to a policy prior, i.e., we regularize in function space. We show that functional regularization yields a bias-variance trade-off, and propose an adaptive tuning strategy to optimize this trade-off. When the policy prior has control-theoretic stability guarantees, we further show that this regularization approximately preserves those stability guarantees throughout learning. We validate our approach empirically on a range of settings, and demonstrate significantly reduced variance, guaranteed dynamic stability, and more efficient learning than deep RL alone.