Saved in:
Bibliographic Details
Main Authors: Simsir, Hikmet, Oguz, Ozgur S.
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2606.01151
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913177037438976
author Simsir, Hikmet
Oguz, Ozgur S.
author_facet Simsir, Hikmet
Oguz, Ozgur S.
contents Behavior cloning with high-capacity generative policies achieves strong imitation performance, but is often limited by demonstration coverage and distribution shift. Direct reinforcement learning fine-tuning can improve performance, but updating large action decoders is frequently unstable and sample inefficient. We propose Lagrangian Perturbation Diffusion Steering (LP-DS), a lightweight adaptation method that improves a frozen generative policy by learning a compact noise-space perturbation before decoding. LP-DS optimizes this perturbation with a Lagrangian trust-region objective, improving downstream value while constraining deviation from the latent prior. Across RoboMimic manipulation, OpenAI Gym locomotion, and Adroit dexterous manipulation benchmarks, LP-DS improves sample efficiency, success, and return while maintaining higher action-space entropy than unconstrained noise-space steering, with return improvements of up to 25% over prior baselines. Additional evaluations with flow-matching backbones, a large vision-language-action model, and physical Franka deployment show that LP-DS is not limited to compact diffusion policies or simulated benchmarks. Project page: https://sites.google.com/view/lp-ds/home.
format Preprint
id arxiv_https___arxiv_org_abs_2606_01151
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Lagrangian Perturbation Diffusion Steering: Latent Reinforcement Learning for Generative Policies
Simsir, Hikmet
Oguz, Ozgur S.
Machine Learning
Behavior cloning with high-capacity generative policies achieves strong imitation performance, but is often limited by demonstration coverage and distribution shift. Direct reinforcement learning fine-tuning can improve performance, but updating large action decoders is frequently unstable and sample inefficient. We propose Lagrangian Perturbation Diffusion Steering (LP-DS), a lightweight adaptation method that improves a frozen generative policy by learning a compact noise-space perturbation before decoding. LP-DS optimizes this perturbation with a Lagrangian trust-region objective, improving downstream value while constraining deviation from the latent prior. Across RoboMimic manipulation, OpenAI Gym locomotion, and Adroit dexterous manipulation benchmarks, LP-DS improves sample efficiency, success, and return while maintaining higher action-space entropy than unconstrained noise-space steering, with return improvements of up to 25% over prior baselines. Additional evaluations with flow-matching backbones, a large vision-language-action model, and physical Franka deployment show that LP-DS is not limited to compact diffusion policies or simulated benchmarks. Project page: https://sites.google.com/view/lp-ds/home.
title Lagrangian Perturbation Diffusion Steering: Latent Reinforcement Learning for Generative Policies
topic Machine Learning
url https://arxiv.org/abs/2606.01151