A Unifying Framework for Action-Conditional Self-Predictive Reinforcement Learning

TL;DR

This paper develops a unified theoretical framework for action-conditional self-predictive reinforcement learning, bridging the gap between existing theory and practical algorithms, and demonstrating improved empirical performance across various settings.

Contribution

It introduces a new action-conditional objective (BYOL-AC), analyzes its convergence, and unifies different objectives through model-based and model-free perspectives.

Findings

BYOL-AC outperforms existing methods in diverse RL environments.

Theoretical analysis reveals convergence properties and relationships between objectives.

Proposes a variance-like objective (BYOL-VAR) with favorable properties.

Abstract

Learning a good representation is a crucial challenge for Reinforcement Learning (RL) agents. Self-predictive learning provides means to jointly learn a latent representation and dynamics model by bootstrapping from future latent representations (BYOL). Recent work has developed theoretical insights into these algorithms by studying a continuous-time ODE model for self-predictive representation learning under the simplifying assumption that the algorithm depends on a fixed policy (BYOL-$\Pi$); this assumption is at odds with practical instantiations of such algorithms, which explicitly condition their predictions on future actions. In this work, we take a step towards bridging the gap between theory and practice by analyzing an action-conditional self-predictive objective (BYOL-AC) using the ODE framework, characterizing its convergence properties and highlighting important distinctions between the limiting solutions of the BYOL-$\Pi$ and BYOL-AC dynamics. We show how the two representations are related by a variance equation. This connection leads to a novel variance-like action-conditional objective (BYOL-VAR) and its corresponding ODE. We unify the study of all three objectives through two complementary lenses; a model-based perspective, where each objective is shown to be equivalent to a low-rank approximation of certain dynamics, and a model-free perspective, which establishes relationships between the objectives and their respective value, Q-value, and advantage function. Our empirical investigations, encompassing both linear function approximation and Deep RL environments, demonstrates that BYOL-AC is better overall in a variety of different settings.