Loop-Free Inverse Reinforcement Learning via Sequential Value Recovery with Q-Score Matching

2026-09-30

AI Summary

This study addresses the high computational overhead and training instability inherent in the bilevel optimization of traditional inverse reinforcement learning (IRL) by proposing a loop-free IRL framework. The method leverages diffusion policies to encode the structure of the optimal soft Q-function, combining action gradient matching with Bellman consistency constraints to reformulate reward learning as sequential value recovery, thereby entirely eliminating the inner policy optimization loop. Furthermore, it introduces Gumbel regression-inspired value estimation and state-dependent offset calibration techniques to enhance robustness. Experimental results demonstrate that the proposed approach achieves a 2–3× training speedup across multiple benchmarks while maintaining or surpassing the reward recovery quality of state-of-the-art methods.

Abstract

Inverse Reinforcement Learning (IRL) aims to recover a reward function that explains expert demonstrations. Existing IRL methods typically rely on a bi-level optimization procedure that alternates between reward learning and policy optimization, leading to substantial computational burden and training instability. In this work, we introduce a different route that eliminates policy optimization entirely by leveraging diffusion policies. Our key insight is that a diffusion policy encodes the action-gradient structure of the optimal soft Q function, enabling reward learning to be cast as a sequence of value recovery problems, thereby allowing us to bypass reward-policy loops inherent in prior IRL methods. Specifically, our method proceeds in three stages: (I) recovering the optimal soft Q function via action-gradient matching and estimating the corresponding soft value function (LogSumExp of Q values) in a way inspired by Gumbel regression; (II) calibrating these soft values by inferring a state-dependent offset; (III) extracting the reward by enforcing Bellman consistency. This leads to Loop-Free Inverse Reinforcement Learning (LFIRL), a fully offline algorithm that operates in a simple, loop-free, and sequential manner. LFIRL is simple to implement and significantly improves training efficiency while maintaining strong reward recovery performance. Empirically, across Maze, Franka Kitchen, Adroit Hand Pen, and Push-T benchmarks, LFIRL achieves 2-3x speedup over the fastest baselines, while matching or surpassing state-of-the-art methods in reward recovery quality.

Problem

• Inverse Reinforcement Learning
• bi-level optimization
• computational burden
• training instability
• reward-policy loop

Innovation

• Inverse Reinforcement Learning
• Diffusion Policy
• Loop-Free Optimization
• Value Recovery
• Q-Score Matching