Bridging the Gap Between Value and Policy Based Reinforcement Learning

TL;DR

This paper introduces Path Consistency Learning (PCL), a novel reinforcement learning algorithm that unifies value and policy-based methods through entropy regularization, demonstrating superior performance on benchmarks.

Contribution

The paper establishes a theoretical link between softmax value consistency and policy optimality, and develops PCL, a new algorithm that combines actor-critic and Q-learning advantages.

Findings

PCL outperforms strong actor-critic and Q-learning baselines.

PCL unifies value and policy-based RL under entropy regularization.

A single model can represent both policy and softmax state values.

Abstract

We establish a new connection between value and policy based reinforcement learning (RL) based on a relationship between softmax temporal value consistency and policy optimality under entropy regularization. Specifically, we show that softmax consistent action values correspond to optimal entropy regularized policy probabilities along any action sequence, regardless of provenance. From this observation, we develop a new RL algorithm, Path Consistency Learning (PCL), that minimizes a notion of soft consistency error along multi-step action sequences extracted from both on- and off-policy traces. We examine the behavior of PCL in different scenarios and show that PCL can be interpreted as generalizing both actor-critic and Q-learning algorithms. We subsequently deepen the relationship by showing how a single model can be used to represent both a policy and the corresponding softmax state values, eliminating the need for a separate critic. The experimental evaluation demonstrates that PCL significantly outperforms strong actor-critic and Q-learning baselines across several benchmarks.