PIPER: Primitive-Informed Preference-based Hierarchical Reinforcement Learning via Hindsight Relabeling

Published: 19 Jun 2024, Last Modified: 26 Jul 2024ARLET 2024 PosterEveryoneRevisionsBibTeXCC BY 4.0
Keywords: hierarchical reinforcement learning, preference learning
TL;DR: We present PIPER, a preference learning-based approach to hierarchical reinforcement learning that addresses the issues of non-stationary rewards and infeasible subgoal selection.
Abstract: In this work, we introduce PIPER: Primitive-Informed Preference-based Hierarchical reinforcement learning via Hindsight Relabeling, a novel approach that leverages preference-based learning to learn a reward model, and subsequently uses this reward model to relabel higher-level replay buffers. Since this reward is unaffected by lower primitive behavior, our relabeling-based approach is able to mitigate non-stationarity, which is common in existing hierarchical approaches, and demonstrates impressive performance across a range of challenging sparse-reward tasks. Since obtaining human feedback is typically impractical, we propose to replace the human-in-the-loop approach with our primitive-in-the-loop approach, which generates feedback using sparse rewards provided by the environment. Moreover, in order to prevent infeasible subgoal prediction and avoid degenerate solutions, we propose primitive-informed regularization that conditions higher-level policies to generate feasible subgoals. We perform extensive experiments to show that PIPER mitigates non-stationarity in hierarchical reinforcement learning and achieves greater than 50$\\%$ success rates in challenging, sparse-reward robotic environments, where most other baselines fail to achieve any significant progress.
Submission Number: 32
Loading