Learning The Minimum Action Distance

TL;DR

This paper introduces a self-supervised framework to learn the minimum action distance (MAD) between states in MDPs solely from state trajectories, enabling improved goal-conditioned tasks without reward signals.

Contribution

It proposes a novel method to learn state representations based on MAD, capturing environment structure without requiring rewards or actions, applicable to various dynamics and observation noise.

Findings

Efficiently learns accurate MAD representations across diverse environments.

Outperforms existing state representation methods in quality.

Works with deterministic, stochastic, discrete, and continuous settings.

Abstract

This paper presents a state representation framework for Markov decision processes (MDPs) that can be learned solely from state trajectories, requiring neither reward signals nor the actions executed by the agent. We propose learning the minimum action distance (MAD), defined as the minimum number of actions required to transition between states, as a fundamental metric that captures the underlying structure of an environment. MAD naturally enables critical downstream tasks such as goal-conditioned reinforcement learning and reward shaping by providing a dense, geometrically meaningful measure of progress. Our self-supervised learning approach constructs an embedding space where the distances between embedded state pairs correspond to their MAD, accommodating both symmetric and asymmetric approximations. We evaluate the framework on a comprehensive suite of environments with known MAD values, encompassing both deterministic and stochastic dynamics, as well as discrete and continuous state spaces, and environments with noisy observations. Empirical results demonstrate that the proposed approach not only efficiently learns accurate MAD representations across these diverse settings but also significantly outperforms existing state representation methods in terms of representation quality.