Saved in:
Bibliographic Details
Main Authors: Wu, Feiyang, Zhao, Ye, Wu, Anqi
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.03013
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918528212271104
author Wu, Feiyang
Zhao, Ye
Wu, Anqi
author_facet Wu, Feiyang
Zhao, Ye
Wu, Anqi
contents We propose a distributional framework for offline Inverse Reinforcement Learning (IRL) that jointly models uncertainty over reward functions and full distributions of returns. Unlike conventional IRL approaches that recover a deterministic reward estimate or match only expected returns, our method captures richer structure in expert behavior, particularly in learning the reward distribution, by minimizing first-order stochastic dominance (FSD) violations and thus integrating distortion risk measures (DRMs) into policy learning, enabling the recovery of both reward distributions and distribution-aware policies. This formulation is well-suited for behavior analysis and risk-aware imitation learning. Theoretical analysis shows that the algorithm converges with $\mathcal{O}(\varepsilon^{-2})$ iteration complexity. Empirical results on synthetic benchmarks, real-world neurobehavioral data, and MuJoCo control tasks demonstrate that our method recovers expressive reward representations and achieves state-of-the-art performance.
format Preprint
id arxiv_https___arxiv_org_abs_2510_03013
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Distributional Inverse Reinforcement Learning
Wu, Feiyang
Zhao, Ye
Wu, Anqi
Machine Learning
We propose a distributional framework for offline Inverse Reinforcement Learning (IRL) that jointly models uncertainty over reward functions and full distributions of returns. Unlike conventional IRL approaches that recover a deterministic reward estimate or match only expected returns, our method captures richer structure in expert behavior, particularly in learning the reward distribution, by minimizing first-order stochastic dominance (FSD) violations and thus integrating distortion risk measures (DRMs) into policy learning, enabling the recovery of both reward distributions and distribution-aware policies. This formulation is well-suited for behavior analysis and risk-aware imitation learning. Theoretical analysis shows that the algorithm converges with $\mathcal{O}(\varepsilon^{-2})$ iteration complexity. Empirical results on synthetic benchmarks, real-world neurobehavioral data, and MuJoCo control tasks demonstrate that our method recovers expressive reward representations and achieves state-of-the-art performance.
title Distributional Inverse Reinforcement Learning
topic Machine Learning
url https://arxiv.org/abs/2510.03013