导读
英文标题：T3R: Training-Free Two-Stage Token Refinement Towards Efficient and Robust VLA Models arXiv编号：2611.08413 作者：Mohammed Hassan, Zhenyang Chen, Zheng…
【编辑部】这对你意味着什么
该论文提出一种无需训练的两阶段 Token 优化方法 T3R，旨在提升视觉-语言-动作模型的推理效率与鲁棒性。实验显示在 LIBERO-Spatial 任务中可移除 85% 的场景相机视觉 token，同时基本保持未剪枝的成功率，并在分布外物体摆放扰动下将成功率提升最多 7 个百分点。推理阶段剪枝与注意力偏置带来的额外开销为每次调用 7.8 ms，随着输入分辨率提高，LLM 前向推理最高可获得 3.3 倍加速。对于关注机器人感知与控制算法落地的华人研究者或工程师，这一方案提供了一种免训练的轻量化部署思路。
英文标题：T3R: Training-Free Two-Stage Token Refinement Towards Efficient and Robust VLA Models
arXiv编号：2611.08413
作者：Mohammed Hassan, Zhenyang Chen, Zheng Wang
来源：arXiv
原文链接：https://arxiv.org/abs/2611.08413
中文摘要：
Vision-Language-Action (VLA) models enable end-to-end robot control, but inference is increasingly expensive. Each decision processes hundreds of visual tokens, often across multiple cameras, leading to substantial computation while allowing irrelevant background content to distract action prediction under distribution shift. We argue that these issues stem from two complementary forms of redundancy: many visual tokens are spatially irrelevant, while attention over the remaining tokens is often misallocated. Based on this observation, we propose T3R, a Training-Free Two-Stage Token Refinement framework. In the first stage, segmentation-guided token pruning removes irrelevant visual patches to reduce computation and eliminate distracting context. In the second stage, saliency-guided attention biasing reallocates attention toward action-critical visual regions and language instructions within the retained tokens. Together, pruning produces a cleaner candidate set, while attention biasing further emphasizes the most decision-relevant evidence, yielding a better accuracy-efficiency frontier than either stage alone. On LIBERO-Spatial, T3R removes 85% of scene-camera visual tokens while essentially matching the unpruned success rate, and improves success by up to 7 percentage points under out-of-distribution object-placement perturbations. The timing comparison attributes 7.8 ms per inference call to pruning and attention-bias application. As input resolution increases, LLM-forward profiling shows up to a 3.3x speedup over the unpruned baseline. Evaluations across LIBERO, SIMPLER, and RoboTwin with three VLA backbones show the strongest accuracy-efficiency benefits in multi-camera settings. These results demonstrate that training-free token refinement is a practical approach to efficient and robust VLA inference.
本文由华人空间 AI 论文助手自动抓取翻译，仅供学术交流。
【编辑部】今日讨论
你在实际部署 VLA 模型时，是否遇到过多摄像头视觉 token 导致推理延迟过高或抗干扰能力不足的问题？
