REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models

Houlong Xiong; Zhenqi Qiu; Zechen Wang; Suohang Zhang; Yiyu Ren; Wanting Xu; Hongfei Niu; Chengyang He

cs.ROOct 8, 2026

REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models

Authors: Houlong Xiong, Zhenqi Qiu, Zechen Wang, Suohang Zhang, Yiyu Ren, Wanting Xu, Hongfei Niu, Chengyang He, +3 more

Organizations: Shanghai Jiao Tong University · PrimeBot · ShanghaiTech University · Zhejiang University · National University of Singapore

Abstract

Flow-based vision-language-action (VLA) models generate action chunks for temporally coherent robot motion, but chunked control creates a fundamental closed-loop trade-off: long chunks provide smooth execution, whereas frequent replanning improves reactivity at the cost of action discontinuities. We introduce REACT, a rolling-denoising framework that makes flow-based VLAs more reactive while preserving long-horizon context. Instead of regenerating entire action chunks from scratch, REACT maintains a persistent action buffer with staggered flow timesteps. At each control step, the full horizon is denoised using the latest observation, the cleanest action block is executed, partially refined future blocks are shifted forward, and fresh noise is appended to the tail. As a result, each executed action block is refined across multiple recent observations before deployment. To support real-time control, we further introduce dual decoupling, which separates sensing, VLM encoding, DiT denoising, and action execution, enabling high-frequency observation updates and action streaming under practical compute constraints. Across the RoboTwin 2.0 simulation benchmark and real-world tasks spanning bimanual manipulation and dynamic control on multiple robot platforms, REACT improves task success and reduces reaction latency while producing smoother trajectories than frequent-replanning and asynchronous baselines.

Figures & tables

Figure 1: Overview of REACT. REACT reformulates flow-based action generation into a receding-horizon rolling denoising process. At each control step, the active horizon is refined conditioned on the latest observation embedding; the oldest refined chunk is executed, while the remaining buffer is shifted forward and replenished with fresh noise. The colored borders track the same action chunk as it is progressively refined across consecutive observations before execution. The right panels summarize throughput and overall performance, showing that REACT reaches the high-input, high-output regime while improving success, reactivity, smoothness, and efficiency.

Figure 2: Real-world general manipulation tasks. ARX X5 (top) and Franka R3 (bottom) perform two shared tasks, Bowl Stacking by Size and Bottle Cap Unscrewing, plus robot-specific hanging tasks: Cable Hanging and Keyring Hanging, respectively.

	$\pi_{0.5}$ H50E50		$\pi_{0.5}$ H50E10		$\pi_{0.5}$ H10E10		Async+RTC		REACT w/o DD		REACT
Task	Clean	Rand	Clean	Rand	Clean	Rand	Clean	Rand	Clean	Rand	Clean	Rand
beat_block_hammer	17	1	8	0	0	0	13	4	15	2	14	4
click_bell	61	6	71	18	35	0	64	8	89	32	61	22
move_playingcard_away	59	26	71	13	7	1	64	11	71	25	66	22
pick_diverse_bottles	20	3	9	0	0	1	10	6	27	9	33	6
press_stapler	58	23	22	3	9	4	37	13	67	42	63	39

Table 1: Simulation results on RoboTwin 2.0 benchmark. Success rate (%) under clean and randomized settings. Best in bold , second best underlined .

Robot	Task	$\pi_{0.5}$ H50E50	$\pi_{0.5}$ H50E10	$\pi_{0.5}$ H10E10	Async+RTC	REACT w/o DD	REACT
ARX X5	Bowl Stacking by Size	73	66	70	36	86	76
	Bottle Cap Unscrewing	36	0	0	0	56	47
	Cable Hanging	56	66	56	3	63	80
	ARX X5 Avg.	55.0	44.0	42.0	13.0	68.3	67.7
Franka	Bowl Stacking by Size	60	50	0	47	66	63
	Bottle Cap Unscrewing	57	0	0	3	63	70

Table 2: Real-world results on general manipulation tasks. Success rate (%) across ARX X5 and Franka platforms. Best in bold , second best underlined .

Figure 3: Trajectory smoothness comparison. Mean jerk norm averaged over successful episodes in simulation, ARX X5, and Franka settings. Demos (GT) applies the same metric to the imitation-learning demonstrations as a reference. Full per-task values are reported in Appendix D .

Panel of Figure 4 — Figure 4: Specialized dynamic closed-loop control tasks. (a) Pour Rice requires the robot to pour rice into a vessel and stop at the target state, testing whether the policy can react to continuously changing flow while maintaining smooth motion. (b) Reaction Game requires the robot to click the mouse when the on-screen signal turns green, directly measuring visual-to-motor reaction latency.

Task	Metric	$\pi_{0.5}$ H50E50	$\pi_{0.5}$ H50E10	$\pi_{0.5}$ H10E10	Async+RTC	REACT w/o DD	REACT	Human Avg.
Pour Rice (ARX X5)	Success (%) $\uparrow$	3	33	0	0	57	63	–
Reaction Game (Franka R3)	Latency (ms) $\downarrow$	1505 $\pm$ 502	768 $\pm$ 113	772 $\pm$ 119	747 $\pm$ 123	754 $\pm$ 127	734 $\pm$ 94	705 $\pm$ 73

Table 3: Specialized task results. Pour Rice reports success rate (%) (higher is better). Reaction Game reports mean

\pm

standard deviation latency in milliseconds (lower is better).

Panel of Figure 5 — Figure 5: Training efficiency test. (a) Training loss over optimization steps. (b) Average closed-loop success rate as a function of training steps; shaded bands show the task-wise range.

Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: Over-pouring under long open-loop execution in Pour Rice. Consecutive frames from the H50E50 baseline show the policy continuing the pouring motion even after the rice level exceeds the target mark, illustrating how stale open-loop execution can overshoot the target state before the next observation update.

Figure 7: Trajectory response in Bowl Stacking by Size when the bowl is suddenly moved during grasping. Left: the original $\pi_{0.5}$ full-chunk policy reacts slowly because it continues executing a stale action chunk. Middle: the shortened action-chunk baseline updates more frequently but produces discontinuous stop-and-replan motion. Right: REACT w/o DD adapts to the moved bowl while maintaining a smoother approach trajectory.

Figure 8: Repeated cap-contact retries under shortened execution in Bottle Cap Unscrewing. Representative frames from the H50E10 baseline show the policy repeatedly returning to the cap-contact region instead of sustaining the continuous grasp–twist–release sequence required to open the bottle.

Figure 9: Dual Decoupling timing diagram. Sensing, VLM encoding, DiT denoising, and action execution run as separate streams. The scheduler selects observations every $\Delta_{S}=S/f_{\mathrm{cam}}$ from the camera stream, and parallel VLM workers encode them into a latest-ready embedding cache; each update consumes the freshest completed embedding and the rolling action buffer, then commits the next $S$ executable actions. The architecture removes both the execution wait and the VLM wait from the DiT critical path: camera-rate input occurs in the $S=1$ setting, while output action throughput reaches the camera cadence whenever the VLM pool and DiT keep up.

Method	$\Phi_{\text{in}}$ (obs/sec)	$\Phi_{\text{out}}$ (act/sec)
Synchronous	$\dfrac{1}{T_{\text{infer}}+H_{e}/f_{\text{ctrl}}}$	$\dfrac{H_{e}}{T_{\text{infer}}+H_{e}/f_{\text{ctrl}}}$
Sync + Accel ( $\alpha$ )	$\dfrac{1}{\alpha T_{\text{infer}}+H_{e}/f_{\text{ctrl}}}$	$\dfrac{H_{e}}{\alpha T_{\text{infer}}+H_{e}/f_{\text{ctrl}}}$
Asynchronous	$\dfrac{1}{\max(T_{\text{infer}},H_{e}/f_{\text{ctrl}})}$	$\dfrac{H_{e}}{\max(T_{\text{infer}},H_{e}/f_{\text{ctrl}})}$
REACT w/o DD	$\dfrac{1}{T_{\text{VLM}}+T_{\text{DiT}}+S/f_{\text{ctrl}}}$	$\dfrac{S}{T_{\text{VLM}}+T_{\text{DiT}}+S/f_{\text{ctrl}}}$
REACT	$\min\left(\frac{f_{\text{cam}}}{S},\frac{1}{T_{\text{DiT}}}\right)$ , if $R_{\text{VLM}}\!\geq\!f_{\text{cam}}/S$	$\min\left(f_{\text{cam}},\frac{S}{T_{\text{DiT}}}\right)$

Appendix

Table 4: Throughput formulas for different inference schemes.

T_{\text{infer}}=T_{\text{VLM}}+N\cdot T_{\text{DiT}}

Method	$H_{e}$ or $S$	$T_{\text{cycle}}$ (ms)	$\Phi_{\text{in}}$ (Hz)	$\Phi_{\text{out}}$ (Hz)
Synchronous	$H_{e}{=}50$	$1734$	$0.58$	$28.8$
Synchronous	$H_{e}{=}25$	$900$	$1.11$	$27.8$
Synchronous	$H_{e}{=}10$	$400$	$2.50$	$25.0$
Synchronous	$H_{e}{=}5$	$234$	$4.28$	$21.4$
Sync + Accel ( $\alpha{=}0.5$ )	$H_{e}{=}50$	$1700$	$0.59$	$29.4$
Sync + Accel ( $\alpha{=}0.5$ )	$H_{e}{=}25$	$867$	$1.15$	$28.8$

Appendix

Table 5: Throughput comparison based on

\pi_{0.5}

(

H{=}50

f_{\text{cam}}{=}30

Hz,

f_{\text{ctrl}}{=}30

Hz,

T_{\text{VLM}}{=}37

ms,

T_{\text{DiT}}{=}3

ms,

N{=}10

Panel of Figure 10 — Figure 10: Hardware platforms. Representative views of the Acone/ARX X5 bimanual platform and the Franka Research 3 dual-arm platform used in the real-robot evaluation.

Hyperparameter	Value
Peak learning rate	$2.5\times 10^{-5}$
Final decay learning rate	$2.5\times 10^{-6}$
LR scheduler	Cosine decay with linear warmup
Warmup fraction	$10\%$ of the total training schedule
Optimizer	AdamW
AdamW $\beta_{1}$	0.9

Appendix

Table 6: Training hyperparameters used for simulation and real-robot policies.

Policy	Latency (ms) $\downarrow$	Speedup	Notes
Standard $\pi_{0.5}$	66.04	1.00 $\times$	VLM embedding + 10 denoising steps.
REACT w/o DD	39.96	1.65 $\times$	VLM embedding + single denoising step.

Appendix

Table 7: Real-robot per-update inference latency. Mean synchronous policy-update latency measured with torch.compile enabled.

Setting	Task	$\pi_{0.5}$ H50E50	$\pi_{0.5}$ H50E10	$\pi_{0.5}$ H10E10	Async+RTC	REACT w/o DD	REACT
Simulation	beat_block_hammer	463.4	996.7	–	1755.0	664.0	775.9
	click_bell	502.4	1087.2	1341.8	1946.4	690.8	712.0
	move_playingcard_away	814.3	2012.7	1784.5	1832.2	1165.8	1341.8
	pick_diverse_bottles	857.7	1391.2	–	1688.0	1253.4	1621.8
	press_stapler	567.9	1295.5	1425.7	3168.2	1136.4	1204.7
	rotate_qrcode	965.5	1663.9	1638.8	1944.5	1296.1	1422.3

Appendix

Table 8: Full trajectory smoothness results. Mean jerk norm of executed joint trajectories for each task and platform; lower indicates smoother motion. Best in bold , second best underlined .

Figure 11: Ablation on block length and block count. Mean RoboTwin clean success rate for each $S{\times}K$ configuration; hatched cells indicate configurations not evaluated.

Figure 12: Per-task ablation heatmaps. Success rate for each of the seven RoboTwin tasks as a function of block length $S$ and block count $K$ . Hatched cells indicate configurations not evaluated. The color scale is shared across all panels (0–100%).

Figure 13: OOD open-loop action prediction. Open-loop action-prediction MSE on the demo_randomized OOD dataset for three representative RoboTwin tasks.

Panel of Figure 14 — Figure 14: Qualitative open-loop predictions on Move Playing Card Away. Two visualizations of the same demo_randomized example show how the predicted action sequence compares with the demonstration trajectory under the baseline and REACT .

Explore similar work

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
RAVEL: Asynchronous Rolling Inference for Flow-Based Vision-Language-Action Models
Sep 28, 2026Yuhan Chen, Ke Yu, Pengfei Liu +3Efficient VLA Model Inference Vision-Language-Action Models
Zhejiang University · Nanyang Technological University
Urgent Actions Go First: Urgency-Aware Denoising for Real-Time VLA Control
Sep 29, 2026Zibo Wang, Haochen Han, Pengzhen Ren +2Robotic Control Efficient VLA Model Inference
Pengcheng Laboratory

Jun 12, 2026, cs.ROJ/K move · Enter open · S save

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

Yanzhao Guo, Wenkai Chen, Jianwei Zhang

Shanghai Jiao Tong University · Technical Aspects of Multimodal Systems (TAMS), Department of Informatics, Universität Hamburg

Language-Conditioned Robot Manipulation Efficient VLM Inference Attention Residual Connections Vision-Language-Action Models

Open paper AskPDF arXiv

Diffusion-based Vision-Language-Action (VLA) policies have demonstrated strong capability in modeling expressive and multimodal action distributions. However, their reliance on iterative sampling introduces substantial inference latency, which limits their applicability to reactive closed-loop robot manipulation. To address this limitation, we propose \texttt{ReactVLA}, a lightweight and low-latency VLA framework for real-time robotic manipulation. \texttt{ReactVLA} combines two complementary designs: (1) an improved Mean Flow (iMF) action generator that reduces expensive multi-step diffusion sampling to one-to-few-step action generation, and (2) Attention Residuals (AttnRes), a dynamic depth-wise feature routing mechanism that replaces uniform residual accumulation to better preserve task-relevant multimodal representations. We evaluate \texttt{ReactVLA} on large-scale simulation benchmarks, including LIBERO and RoboIMI, as well as real-world robotic manipulation tasks. Experimental results show that \texttt{ReactVLA} consistently outperforms similarly sized VLA baselines, including SmolVLA and

π_0

. On challenging precision manipulation tasks, \texttt{ReactVLA} achieves up to a 1.65

\times

improvement in task performance while providing more than a 4

\times

increase in inference speed compared with leading VLA models. Finally, it reduces real-world policy latency to below 38.6 ms, enabling fast reactive control on physical robot platforms. Please check out our project website at: https://game-loader.github.io/ReactVLA/.

Abstract

Explore similar work

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

RAVEL: Asynchronous Rolling Inference for Flow-Based Vision-Language-Action Models

Urgent Actions Go First: Urgency-Aware Denoising for Real-Time VLA Control