High-Fidelity One-Step Generative Visuomotor Policy via Recursive Correction - Frequency Consistency - Contrastive Flow Matching

Sep 21, 2026

Uploaded by Asher

Generative visuomotor policies driven by diffusion and flow matching model complex, multimodal action distributions, yet multi-step sampling and ODE solvers create significant inference latency. Single-step accelerations often compress the iterative generation into one update, yielding an approximation gap reflected in spatial deviation, frequency distortion, and mode averaging. A high-fidelity one-step framework is proposed using recursive correction, dual-timestep frequency consistency, and contrastive flow matching, producing competitive or superior manipulation results with only a single forward pass (1 NFE).

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Abstract

Introduction

Visual imitation learning for visuomotor control

Generative continuous-time policies and their latency challenge

Single-step acceleration and the approximation gap

High-Fidelity One-Step Generative Visuomotor Policy via Recursive Correction, Frequency Consistency, and Contrastive Flow Matching Yuran Chen, Xinye Cai, Zhonglin Gong, and Yang Huang arXiv :2607 .03865v 1 [ cs .RO] 4 Jul 2026 Abstract—Generative models such as Diffusion and Flow Matching have significantly advanced robotic visuomotor policies by modeling complex and multimodal action distributions. However, their reliance on multi-step sampling or ODE solvers introduces substantial inference latency. Existing singlestep acceleration methods improve efficiency, but they often approximate the entire iterative action-generation process with a single large update, introducing an approximation gap. We characterize this gap as three coupled failure modes: spatial deviation, where large-step approximation ignores local flow geometry; frequency distortion, where over-smoothed predictions suppress fine-grained manipulation details; and mode averaging, where entangled multimodal flows produce ambiguous actions. To address these challenges, we propose a high-fidelity one-step generative visuomotor policy framework based on three complementary mechanisms: recursive correction, frequency consistency, and contrastive flow matching. First, Recursive Consistent Action Flow (RCAF) performs high-order recursive correction to compensate for spatial truncation errors, aligning one-step predictions with recursively refined flow trajectories. Second, Dual-Timestep Frequency Consistency (DTFC) preserves highfrequency manipulation details by encouraging adaptive spectral consistency across different flow timesteps. Third, Contrastive Flow Matching (CFM) separates entangled action flows with a margin-based repulsive objective, reducing mode averaging and action ambiguity in multimodal manipulation scenarios. Extensive experiments on RoboTwin, RoboTwin 2.0, Adroit, DexArt, and real-world robot platforms show that our method achieves competitive or superior manipulation performance compared with strong 10-step generative policy baselines while requiring only a single forward pass (1 NFE), reducing inference cost for low-latency visuomotor control. Index Terms—Visuomotor Policy, Flow Matching, Consistency Model, Robot Learning I. INTRODUCTION Visual imitation learning [1], [2] enables robots to acquire manipulation skills by learning direct mappings from observations to actions from expert demonstrations. As an important paradigm in robot learning, it has shown promising results ina wide range of manipulation scenarios, including dexterous grasping, precise assembly, bimanual coordination, and longhorizon object manipulation [3], [4], [5], [6], [7], [8], [9],[10] . This work was supported in part by the National Natural Science Foundation of China under Grants 62572005 and U24A20267 . (Corresponding author: Xinye Cai.) Yuran Chen, Zhonglin Gong, and Yang Huang are with the School of Safety Science and Engineering, Anhui University of Science and Technology, Huainan 232001, China, and also with the State Key Laboratory of Digital Intelligent Technology for Unmanned Coal Mining, Anhui University of Science and Technology, Huainan 232001, China (e-mail: [1014913959@qq.com](1014913959@qq.com)). Xinye Cai is with the State Key Laboratory of Digital Intelligent Technology for Unmanned Coal Mining, Anhui University of Science and Technology, Huainan 232001, China, and also with the School of Artificial Intelligence, Anhui University of Science and Technology, Hefei 231131, China (e-mail: [xinye@aust.edu.cn](xinye@aust.edu.cn)). A central challenge in this setting is that expert action distributions are often high-dimensional, temporally correlated, and multimodal, since multiple valid action sequences may solve the same task under the same visual observation. Recently, generative models based on continuous-time dynamics, such as diffusion models [11],[12] and flow matching models [13], [14], have become effective tools for visuomotor poli

FAQs

What approximation gap do current single-step generative methods introduce?

They compress an iterative action-generation process into one large update, leading to spatial deviation, frequency distortion, and mode averaging.

How does the proposed method address spatial deviation?

It uses Recursive Consistent Action Flow (RCAF) for high-order recursive correction that aligns one-step predictions with refined flow trajectories.

How many inference evaluations does the proposed policy require and why is it important?

It needs only a single forward pass (1 NFE), lowering inference cost and enabling low-latency closed-loop visuomotor control.

Sep 21, 2026

Uploaded by Asher