具身开发者社区

外观

论文arXiv 2610.0813322 天前

VLA-ACL：用于高效 Vision-Language-Action 模型的动作一致视觉 Token 剪枝 ​

针对VLA实时部署中长token计算开销，VLA-ACL以动作一致性学习轻量视觉token剪枝，并在LIBERO和真实操作任务验证。

AI 整理
  • 本体 SDK 与控制
  • 具身模型
  • 端侧部署

原文链接：VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models（arXiv 2610.08133 · 发布日期：2026-10-06）

VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models

针对VLA实时部署中长token计算开销，VLA-ACL以动作一致性学习轻量视觉token剪枝，并在LIBERO和真实操作任务验证。

核心要点 ​

  • VLA模型每个控制步处理长token序列，视觉patch冗余导致计算成本高。
  • 方法冻结基础VLA，通过动作级监督学习轻量视觉token剪枝策略。
  • 训练鼓励剪枝上下文动作与全上下文教师一致，并辅以真实动作监督。
  • 在LIBERO和真实操作任务中，最多剪除87.5%视觉token、计算降75%、推理提速1.5倍；代码已开源。

验证范围与局限：验证范围为LIBERO和真实世界操作任务。

英文摘要 (Abstract) ​

Vision-Language-Action (VLA) models achieve strong robotic manipulation performance but incur high computational costs from processing long token sequences at every control step, limiting real-time deployment. Visual token pruning offers a direct solution, as visual patches dominate the input sequence and contain considerable redundancy. Existing approaches, however, either rely on indirect training-free heuristics, such as attention scores and motion thresholds, or require costly fine-tuning of the base VLA model. We introduce VLA-ACL (Action Consistency Learning), which learns a lightweight visual token pruning policy through action-level supervision while keeping the base VLA model entirely frozen. The training objective encourages actions produced from pruned visual contexts to remain consistent with the full-context teacher, with ground-truth actions as auxiliary supervision. This directly ties token selection to its effect on the downstream control output. Experiments on LIBERO and real-world manipulation tasks show that VLA-ACL prunes up to 87.5% of visual tokens while retaining competitive performance, reduces computation by up to 75%, and achieves a 1.5x inference speedup. These results establish a stronger performance-efficiency trade-off than existing frozen-VLA pruning methods and demonstrate the value of action-level supervision for visual token selection. Code is available at https://github.com/du-owen/VLA-ACL.

作者：Owen Du, Yang Yue, Jie Zhang, Jiaqi Pi, Chongxue Chen, Gao Huang

来源

本页事实对应的公开页面，以原文为准。

  1. arXiv 2610.08133：VLA-ACL: Action-Consistent Visual Token Pruning for Efficient Vision-Language-Action Models访问 2026-10-21