Action Robust Reinforcement Learning and Applications in Continuous Control

TL;DR

This paper introduces new robustness criteria for reinforcement learning policies against action uncertainties, develops algorithms for these criteria, and demonstrates their effectiveness and regularization benefits in continuous control tasks.

Contribution

It formalizes two novel robustness criteria for action uncertainty, proposes algorithms for these criteria, and extends them to deep RL with successful experiments in MuJoCo environments.

Findings

Robust policies withstand adversarial action perturbations.

Action robustness improves performance even without perturbations.

Proposed methods act as implicit regularizers in RL.

Abstract

A policy is said to be robust if it maximizes the reward while considering a bad, or even adversarial, model. In this work we formalize two new criteria of robustness to action uncertainty. Specifically, we consider two scenarios in which the agent attempts to perform an action $a$, and (i) with probability $\alpha$, an alternative adversarial action $\bar a$ is taken, or (ii) an adversary adds a perturbation to the selected action in the case of continuous action space. We show that our criteria are related to common forms of uncertainty in robotics domains, such as the occurrence of abrupt forces, and suggest algorithms in the tabular case. Building on the suggested algorithms, we generalize our approach to deep reinforcement learning (DRL) and provide extensive experiments in the various MuJoCo domains. Our experiments show that not only does our approach produce robust policies, but it also improves the performance in the absence of perturbations. This generalization indicates that action-robustness can be thought of as implicit regularization in RL problems.