VLA-Rolling: Fewer Denoising Steps, Better Control via Adaptive Action Reuse
Summary
VLA-Rolling is a lightweight inference framework that reuses discarded action tails from flow-matching VLA models with adaptive denoising budgets, achieving up to 1.48x speedup and up to 15.08 percentage-point success rate gains across 16 model-benchmark pairs on LIBERO, MetaWorld, LIBERO-Plus, and RoboTwin 2.0.
View Cached Full Text
Cached at: 10/21/26, 07:02 AM
Paper page - VLA-Rolling: Fewer Denoising Steps, Better Control via Adaptive Action Reuse
Source: https://huggingface.co/papers/2610.25043 Authors:
,
,
,
,
,
,
,
,
,
,
,
Abstract
VLA-Rolling improves vision-language-action model efficiency by reusing discarded action tails with adaptive denoising budgets, reducing inference cost and boosting success rates.
Flow-matchingvision-language-action (VLA) models generate anaction chunkby iterativelydenoisingGaussian noise, but typically execute only a short prefix before replanning with updated observations. Under thisreceding-horizonparadigm, the unexecuted remainder---the action tail---is discarded, and the next chunk is regenerated from fresh noise. We show that this discarded tail retains transferabledenoisingprogress:action-tail initializationcan match or outperform a native 10-step Gaussian baseline using fewerdenoisingsteps. However, the optimaldenoisingbudget varies substantially across samples, indicating that tail reuse should be adapted to the updated environment. Based on this observation, we introduceVLA-Rolling, a lightweight inference framework that reuses action tails through three stages: action-tail inheritance, per-actionconfidence estimation, and confidence-guided refinement throughrenoisinganddenoisingbudget allocation, while the underlying VLA remains frozen. Across π_{0.5}, SmolVLA, X-VLA, and MolmoAct2 on LIBERO, MetaWorld, LIBERO-Plus, and RoboTwin 2.0,VLA-Rollingreduces action-chunk inference cost in all 16 model--benchmark pairs, achieving up to a 1.48times speedup, while improving success rate in 14 pairs by up to 15.08 percentage points. These results show that action tails are not merely discarded predictions, but reusable intermediate states that enable more efficient and accurate VLA inference. These results demonstrate that the action tail can support the generation of better actions with fewerdenoisingsteps. Our project is available at https://chenbao123ww.github.io/VLA-Rolling/.
View arXiv pageView PDFProject pageGitHub6Add to collection
Get this paper in your agent:
hf papers read 2610\.25043
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper1
#### Chenbao77/VLA-Rolling Robotics• Updatedabout 5 hours ago
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2610.25043 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2610.25043 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
RL^2-VLA: Adaptive RL Latent Compositional Steering with Test-Time Scaling for Vision-Language-Action Models
This paper introduces RL^2, an adaptive inference-time steering framework for Vision-Language-Action models that uses offline RL on latent representations to compose action flows, activating steering only when failure is predicted. It achieves up to +17.3% success rate improvements on SIMPLER and PolaRiS benchmarks and demonstrates real-world transfer.
D-VLA: A High-Concurrency Distributed Asynchronous Reinforcement Learning Framework for Vision-Language-Action Models
D-VLA proposes a high-concurrency distributed asynchronous reinforcement learning framework for Vision-Language-Action models, using plane decoupling and a swimlane pipeline to improve throughput and efficiency in large-scale embodied AI training.
From Foundation to Application: Improving VLA Models in Practice
This paper presents LingBot-VLA 2.0, which enhances VLA foundation models for robotics by improving generalization across tasks and embodiments, expanding action space to whole-body degrees of freedom, and incorporating predictive dynamics modeling for better temporal reasoning.
AR-VLA: True Autoregressive Action Expert for Vision-Language-Action Models
Proposes AR-VLA, an autoregressive action expert that generates continuous action sequences with long-term memory for context-aware robotic policy training, improving trajectory smoothness and task success rates over reactive VLA models.
Early Verdicts, Better Budgets: Sequential Adaptive Rollout Allocation for Compute-Efficient RLVR
This paper introduces SARA, a sequential adaptive rollout allocation method for RLVR that abandons saturated groups early and reallocates the budget, achieving comparable accuracy with 22% fewer rollouts than dynamic sampling and up to 67% savings when combined.