Bad Habits: Policy Confounding and Out-of-Trajectory Generalization in RL

TL;DR

This paper investigates how reinforcement learning agents develop environment-specific habits that can lead to poor generalization when policies change, due to spurious correlations formed during repetitive experiences.

Contribution

It introduces a formal framework for understanding policy confounding and demonstrates its impact on out-of-trajectory generalization in RL.

Findings

Policy confounding causes agents to rely on spurious correlations.

Habits formed during training can mislead agents in new environments.

Mathematical characterization of policy confounding is provided.

Abstract

Reinforcement learning agents tend to develop habits that are effective only under specific policies. Following an initial exploration phase where agents try out different actions, they eventually converge onto a particular policy. As this occurs, the distribution over state-action trajectories becomes narrower, leading agents to repeatedly experience the same transitions. This repetitive exposure fosters spurious correlations between certain observations and rewards. Agents may then pick up on these correlations and develop simplistic habits tailored to the specific set of trajectories dictated by their policy. The problem is that these habits may yield incorrect outcomes when agents are forced to deviate from their typical trajectories, prompted by changes in the environment. This paper presents a mathematical characterization of this phenomenon, termed policy confounding, and illustrates, through a series of examples, the circumstances under which it occurs.