Hierarchical Average-Reward Linearly-solvable Markov Decision Processes

TL;DR

This paper presents a hierarchical reinforcement learning method for LMDPs that efficiently learns low- and high-level tasks simultaneously, significantly outperforming flat RL approaches in average-reward settings.

Contribution

It introduces a novel hierarchical approach that leverages state space partitions and task compositionality for efficient learning in LMDPs.

Findings

Outperforms flat RL by one or more orders of magnitude

Enables simultaneous learning of multiple task levels

Uses state space partitions for easier subtask solutions

Abstract

We introduce a novel approach to hierarchical reinforcement learning for Linearly-solvable Markov Decision Processes (LMDPs) in the infinite-horizon average-reward setting. Unlike previous work, our approach allows learning low-level and high-level tasks simultaneously, without imposing limiting restrictions on the low-level tasks. Our method relies on partitions of the state space that create smaller subtasks that are easier to solve, and the equivalence between such partitions to learn more efficiently. We then exploit the compositionality of low-level tasks to exactly represent the value function of the high-level task. Experiments show that our approach can outperform flat average-reward reinforcement learning by one or several orders of magnitude.