If MaxEnt RL is the Answer, What is the Question?

TL;DR

This paper explores the theoretical foundations of MaxEnt RL, showing it effectively solves certain control problems with reward variability, especially under uncertainty in task goals, and connects it to robust control and POMDPs.

Contribution

The paper formally demonstrates that MaxEnt RL can optimally solve specific classes of control problems with reward uncertainty and links it to robust control and game-theoretic frameworks.

Findings

MaxEnt RL can solve certain POMDPs.

MaxEnt RL is equivalent to a two-player game with an adversarial reward.

Domains with reward uncertainty are well-suited for MaxEnt RL.

Abstract

Experimentally, it has been observed that humans and animals often make decisions that do not maximize their expected utility, but rather choose outcomes randomly, with probability proportional to expected utility. Probability matching, as this strategy is called, is equivalent to maximum entropy reinforcement learning (MaxEnt RL). However, MaxEnt RL does not optimize expected utility. In this paper, we formally show that MaxEnt RL does optimally solve certain classes of control problems with variability in the reward function. In particular, we show (1) that MaxEnt RL can be used to solve a certain class of POMDPs, and (2) that MaxEnt RL is equivalent to a two-player game where an adversary chooses the reward function. These results suggest a deeper connection between MaxEnt RL, robust control, and POMDPs, and provide insight for the types of problems for which we might expect MaxEnt RL to produce effective solutions. Specifically, our results suggest that domains with uncertainty in the task goal may be especially well-suited for MaxEnt RL methods.