A Differential Perspective on Distributional Reinforcement Learning

2025/06/03 by Juan Sebastián Rojas, Rojas, Juan Sebastian, Chi-Guhn Lee +1 · 1 citation
Computer Science · Decision Sciences · #Reinforcement Learning in Robotics #Advanced Bandit Algorithms Research #Adaptive Dynamic Programming Control

paper · pdf · doi:10.48550/arxiv.2506.03333

Abstract

To date, distributional reinforcement learning (distributional RL) methods have exclusively focused on the discounted setting, where an agent aims to optimize a discounted sum of rewards over time. In this work, we extend distributional RL to the average-reward setting, where an agent aims to optimize the reward received per time step. In particular, we utilize a quantile-based approach to develop the first set of algorithms that can successfully learn and/or optimize the long-run per-step reward distribution, as well as the differential return distribution of an average-reward MDP. We derive proven-convergent tabular algorithms for both prediction and control, as well as a broader family of algorithms that have appealing scaling properties. Empirically, we find that these algorithms yield competitive and sometimes superior performance when compared to their non-distributional equivalents, while also capturing rich information about the long-run per-step reward and differential return distributions.

Cited by

• Ergodic Risk Measures: Towards a Risk-Aware Foundation for Continual Reinforcement Learning

Related

• Distributional Reinforcement Learning with Quantile Regression
• Foundations of Multivariate Distributional Reinforcement Learning
• On Policy Evaluation Algorithms in Distributional Reinforcement Learning
• Normality-Guided Distributional Reinforcement Learning for Continuous Control
• Distributional Reinforcement Learning for Multi-Dimensional Reward\n Functions