Epigraph-Guided Flow Matching for Safe and Performant Offline Reinforcement Learning

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Tayal, Manan, Tayal, Mumuksh
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908820545994752
author Tayal, Manan
Tayal, Mumuksh
author_facet Tayal, Manan
Tayal, Mumuksh
contents Offline reinforcement learning (RL) provides a compelling paradigm for training autonomous systems without the risks of online exploration, particularly in safety-critical domains. However, jointly achieving strong safety and performance from fixed datasets remains challenging. Existing safe offline RL methods often rely on soft constraints that allow violations, introduce excessive conservatism, or struggle to balance safety, reward optimization, and adherence to the data distribution. To address this, we propose Epigraph-Guided Flow Matching (EpiFlow), a framework that formulates safe offline RL as a state-constrained optimal control problem to co-optimize safety and performance. We learn a feasibility value function derived from an epigraph reformulation of the optimal control problem, thereby avoiding the decoupled objectives or post-hoc filtering common in prior work. Policies are synthesized by reweighting the behavior distribution based on this epigraph value function and fitting a generative policy via flow matching, enabling efficient, distribution-consistent sampling. Across various safety-critical tasks, including Safety-Gymnasium benchmarks, EpiFlow achieves competitive returns with near-zero empirical safety violations, demonstrating the effectiveness of epigraph-guided policy synthesis.
format Preprint
id arxiv_https___arxiv_org_abs_2602_08054
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Epigraph-Guided Flow Matching for Safe and Performant Offline Reinforcement Learning
Tayal, Manan
Tayal, Mumuksh
Machine Learning
Artificial Intelligence
Offline reinforcement learning (RL) provides a compelling paradigm for training autonomous systems without the risks of online exploration, particularly in safety-critical domains. However, jointly achieving strong safety and performance from fixed datasets remains challenging. Existing safe offline RL methods often rely on soft constraints that allow violations, introduce excessive conservatism, or struggle to balance safety, reward optimization, and adherence to the data distribution. To address this, we propose Epigraph-Guided Flow Matching (EpiFlow), a framework that formulates safe offline RL as a state-constrained optimal control problem to co-optimize safety and performance. We learn a feasibility value function derived from an epigraph reformulation of the optimal control problem, thereby avoiding the decoupled objectives or post-hoc filtering common in prior work. Policies are synthesized by reweighting the behavior distribution based on this epigraph value function and fitting a generative policy via flow matching, enabling efficient, distribution-consistent sampling. Across various safety-critical tasks, including Safety-Gymnasium benchmarks, EpiFlow achieves competitive returns with near-zero empirical safety violations, demonstrating the effectiveness of epigraph-guided policy synthesis.
title Epigraph-Guided Flow Matching for Safe and Performant Offline Reinforcement Learning
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2602.08054