There Is No Turning Back: A Self-Supervised Approach for Reversibility-Aware Reinforcement Learning

TL;DR

This paper introduces a self-supervised method to distinguish reversible from irreversible actions in reinforcement learning, enabling agents to improve exploration and control by leveraging learned reversibility without prior knowledge.

Contribution

The authors present a novel self-supervised approach to estimate action reversibility and incorporate it into RL agents for better decision-making and reduced side-effects.

Findings

Reversibility can be learned via a simple ranking task.

Reversibility-aware agents outperform standard agents in complex environments.

Agents can achieve zero side-effects without reward access.

Abstract

We propose to learn to distinguish reversible from irreversible actions for better informed decision-making in Reinforcement Learning (RL). From theoretical considerations, we show that approximate reversibility can be learned through a simple surrogate task: ranking randomly sampled trajectory events in chronological order. Intuitively, pairs of events that are always observed in the same order are likely to be separated by an irreversible sequence of actions. Conveniently, learning the temporal order of events can be done in a fully self-supervised way, which we use to estimate the reversibility of actions from experience, without any priors. We propose two different strategies that incorporate reversibility in RL agents, one strategy for exploration (RAE) and one strategy for control (RAC). We demonstrate the potential of reversibility-aware agents in several environments, including the challenging Sokoban game. In synthetic tasks, we show that we can learn control policies that never fail and reduce to zero the side-effects of interactions, even without access to the reward function.