How RL Agents Behave When Their Actions Are Modified

TL;DR

This paper introduces the Modified-Action Markov Decision Process to analyze how reinforcement learning agents adapt when their actions are altered by supervisors, revealing different behaviors of RL algorithms in such scenarios.

Contribution

It extends the MDP model to include action modifications and analyzes how various RL algorithms respond, guiding better control of agent behavior under supervision.

Findings

Some RL algorithms ignore action modifications entirely.

Others adapt to avoid modifications that reduce reward.

Proper algorithm choice can prevent agents from circumventing constraints.

Abstract

Reinforcement learning in complex environments may require supervision to prevent the agent from attempting dangerous actions. As a result of supervisor intervention, the executed action may differ from the action specified by the policy. How does this affect learning? We present the Modified-Action Markov Decision Process, an extension of the MDP model that allows actions to differ from the policy. We analyze the asymptotic behaviours of common reinforcement learning algorithms in this setting and show that they adapt in different ways: some completely ignore modifications while others go to various lengths in trying to avoid action modifications that decrease reward. By choosing the right algorithm, developers can prevent their agents from learning to circumvent interruptions or constraints, and better control agent responses to other kinds of action modification, like self-damage.