Mean-Flow based One-Step Vision-Language-Action

Yang Chen; Xiaoguang Ma; Bin Zhao

cs.ROMar 2, 2026

Mean-Flow based One-Step Vision-Language-Action

Authors: Yang Chen, Xiaoguang Ma, Bin Zhao

Abstract

Recent advances in FlowMatching-based Vision-Language-Action (VLA) frameworks have demonstrated remarkable advantages in generating high-frequency action chunks, particularly for highly dexterous robotic manipulation tasks. Despite these notable achievements, their practical applications are constrained by prolonged generation latency, which stems from inherent iterative sampling requirements and architectural limitations. To address this critical bottleneck, we propose a Mean-Flow based One-Step VLA approach. Specifically, we resolve the noise-induced issues in the action generation process, thereby eliminating the consistency constraints inherent to conventional Flow-Matching methods. This significantly enhances generation efficiency and enables one-step action generation. Real-world robotic experiments show that the generation speed of the proposed Mean-Flow based One-Step VLA is 8.7 times and 83.9 times faster than that of SmolVLA and Diffusion Policy, respectively. These results elucidate its great potential as a high-efficiency backbone for VLA-based robotic manipulation.

Figures & tables

Fig. 1: Overview of the Mean-Flow based One-Step VLA framework. The pretrained VLM processes multimodal inputs. During training, the Mean-Flow action expert approximates the mean denoising vector field conditioned on VLM features and predicts the mean vector field to obtain the Action $A_{0}$ during inference.

Fig. 2: Illustrations of Real-world robotic arm and three real manipulation tasks.

Fig. 3: MeanFlow under various $flow\text{-}ratio$ .

Ratio	Success Rates
0.2	84.5
0.5	80.5
1.0	4.5

TABLE I: Success rates (%) under various flow ratios (NFE = 5).

Gamma	Success Rates
0.3	79.5
0.5	86
1.0	9.5

TABLE II: Success rates (%) under various Gamma value with

flow\text{-}ratio

=0.2 (NFE = 5).

NFE	1	2	3	4	5
Success rates	49	48	51.5	79	84.5
NFE	6	7	8	9	10
Success rates	83.5	83.5	86	84.5	85

TABLE III: Success rates (%) under various NFE values with

flow\text{-}ratio

=0.2, Gamma=0.5.

Chunk Size	Success Rates
1	0
10	82
20	84.25
50	50.5

TABLE IV: Success rates (%) under various action chunk size with

flow\text{-}ratio

=0.2, Gamma=0.5 (NFE = 1).

Fig. 4: Illustrations showing the average action-generation speed of Diffusion Policy, SmolVLA, and One-Step VLA across three real manipulation tasks. The One-Step VLA is 8.7 times faster than SmolVLA and 83.9 times faster than Diffusion Policy.

Single-task Training
Policy	PickPlace	Stacking	Sorting	Avg.
Diffusion Policy	78.5	46	60	61.5
Multi-task Training
SmolVLA	86.5	81.5	85.5	84.5
One-Step VLA	88	64	82	78

TABLE V: Success rates (%) of three models evaluated across three different tasks.

Explore similar work

DriftingVLA: Native One-Step Vision-Language-Action Generation via Per-Dimension Temporal Drifting
Aug 30, 2026Yuxuan Gao, Shiqi Zhang, Yedong Shen +6Efficient VLA Model Inference Visuomotor Policy Learning
University of Science and Technology of China, Hefei 230026, China. · LYNSENSE, Shanghai, China.
SnapFlow: One-Step Action Generation for Flow-Matching VLAs via Progressive Self-Distillation
Apr 7, 2026Wuyang Luan, Junhui Li, Weiguang Zhao +3Efficient VLA Model Inference Vision-Language-Action Models
Jilin University · Chongqing University · University of Liverpool +1
ReactVLA: Fast and Lightweight Reactive Robot Manipulation via Improved Mean Flow Action Generation
Jun 12, 2026Yanzhao Guo, Wenkai Chen, Jianwei ZhangLanguage-Conditioned Robot Manipulation Efficient VLM Inference
Shanghai Jiao Tong University · Technical Aspects of Multimodal Systems (TAMS), Department of Informatics, Universität Hamburg

Aug 30, 2026J/K move · Enter open · S save

DriftingVLA: Native One-Step Vision-Language-Action Generation via Per-Dimension Temporal Drifting

Yuxuan Gao, Shiqi Zhang, Yedong Shen, Yifan Duan+5

University of Science and Technology of China, Hefei 230026, China. · LYNSENSE, Shanghai, China.

Efficient VLA Model Inference Visuomotor Policy Learning Efficient VLA Models Vision-Language-Action Models Action Chunking

Open paper AskPDF arXiv

Conventional flow-based vision-language-action (VLA) models support expressive continuous action generation but rely on multi-step refinement to produce each action chunk, increasing latency in online robot control. To address this issue, we introduce DriftingVLA, a native one-step VLA that generates a complete action chunk with a single action-expert forward pass. Rather than learning a flow field that requires iterative integration at inference, DriftingVLA uses a distribution-drifting objective to learn a direct noise-to-action-chunk mapping for one-step deployment. Since robot action dimensions carry distinct control semantics and distributional characteristics, we further introduce Per-Dimension Temporal Drifting (PDTD). PDTD treats the complete temporal trajectory of each action dimension as a separate drifting unit, enabling finer-grained modeling and shaping of dimension-specific action distributions. This per-dimension decomposition applies only to the training objective; the shared VLA model still generates the complete action chunk jointly, thereby preserving cross-dimensional dependencies. DriftingVLA achieves 98.32% success on LIBERO, 81.09% on RoboTwin 2.0, and 77.67% across six real-world single- and dual-arm tasks, outperforming the evaluated multi-step flow policy and one-step VLA baselines. Native one-step deployment also delivers a 3.36-fold speedup in action-chunk generation, eliminating iterative refinement without sacrificing control performance.

Abstract

Explore similar work

DriftingVLA: Native One-Step Vision-Language-Action Generation via Per-Dimension Temporal Drifting

SnapFlow: One-Step Action Generation for Flow-Matching VLAs via Progressive Self-Distillation

ReactVLA: Fast and Lightweight Reactive Robot Manipulation via Improved Mean Flow Action Generation