VLA-Roll, Fewer Denoising Steps, Better Control via Adaptive Action Reuse
VLA-Rolling: Fewer Denoising Steps, Better Control via Adaptive Action Reuse
This research proposes VLA-Rolling, a framework that reuses action tails from vision-language-action models to enhance control efficiency. By adapting denoising budgets and refining actions through confidence-guided steps, the method achieves faster inference and improved success rates across multiple benchmarks.
This paper introduces VLA-Rolling, a lightweight inference framework for vision-language-action models that reuses action tails to improve efficiency and accuracy. The method reduces inference cost across multiple benchmarks and shows that discarded action tails can be reused for better control with fewer denoising steps.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
denoising steps
10
View original evidence
at this discarded tail retains transferable denoising progress: action-tail initialization can match or outperform a native 10-step Gaussian baseline using fewer denoising steps. However, the optimal denoising budget varies substantially across samples, indicating that tail reuse should be adapted to the updated environment. Based on this observation, we introduce VLA-Rolling, a lightweight inference framework that r
Open source S3model--benchmark pairs
16
View original evidence
euses action tails through three stages: action-tail inheritance, per-action confidence estimation, and confidence-guided refinement through renoising and denoising budget allocation, while the underlying VLA remains frozen. Across $\pi_{0.5}$, SmolVLA, X-VLA, and MolmoAct2 on LIBERO, MetaWorld, LIBERO-Plus, and RoboTwin 2.0, VLA-Rolling reduces action-chunk inference cost in all 16 model--benchmark pairs, achieving
Open source S4
- paper: not_reported
- code: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
euses action tails through three stages: action-tail inheritance, per-action confidence estimation, and confidence-guided refinement through renoising and denoising budget allocation, while the underlying VLA remains frozen. Across $\pi_{0.5}$, SmolVLA, X-VLA, and MolmoAct2 on LIBERO, MetaWorld, LIBERO-Plus, and RoboTwin 2.0, VLA-Rolling reduces action-chunk inference cost in all 16 model--benchmark pairs, achieving
Open source S4
up to a $1.48\times$ 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 fewer denoising steps. Our project is available at ht
Open source S5
at this discarded tail retains transferable denoising progress: action-tail initialization can match or outperform a native 10-step Gaussian baseline using fewer denoising steps. However, the optimal denoising budget varies substantially across samples, indicating that tail reuse should be adapted to the updated environment. Based on this observation, we introduce VLA-Rolling, a lightweight inference framework that r
Open source S3
tps://chenbao123ww.github.io/VLA-Rolling/.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org