[论文] GeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Traject...
研究领域: ML 作者: Xin Chen, Sen Chen, Yujuan Ding, Jian Liu, Guoqing Wang, Wei Ye, Heng Tao Shen, Yi Bin 发布时间: 2026-09-17 arXiv: 2609.20776
论文概要
研究领域: ML 作者: Xin Chen, Sen Chen, Yujuan Ding, Jian Liu, Guoqing Wang, Wei Ye, Heng Tao Shen, Yi Bin 发布时间: 2026-09-17 arXiv: 2609.20776
中文摘要
动作分块(action chunking)广泛用于视觉-语言-动作(VLA)策略中的动作生成和执行,但现有方法通常使用固定的动作视界(horizon)。在 rollout 过程中,不同任务阶段可能需要不同的动作连续性、控制精度和闭环反馈水平,使得固定视界无法适应变化的控制需求。我们提出 GeoAAC——一种基于几何的自适应动作分块方法,用于基于流的 VLA 策略,根据当前动作预测的可靠性调整动作视界。我们表明,流匹配(Flow Matching)去噪轨迹的几何结构提供了过程级信息来表征预测可靠性,动作前缀之间的几何变化与预测不确定性保持正相关。GeoAAC 利用这种前缀级几何构建视界级几何剖面,从单次生成中自适应确定动作视界,无需额外训练。在 GR00T N1.5 和 π0.5 上于 LIBERO、LIBERO-Pro、RoboCasa365 和真实世界操控任务上的实验表明,该方法一致优于固定动作视界基线和现有自适应方法,仿真中提升高达 8.7 个百分点,真实世界平均成功率从 53.3% 提高到 74.4%。
原文摘要
Action chunking is widely used for action generation and execution in Vision-Language-Action (VLA) policies, yet existing approaches commonly use a fixed action horizon. During a rollout, different task stages may require different levels of action continuity, control precision, and closed-loop feedback, making a fixed horizon unable to accommodate changing control requirements. We propose GeoAAC, a geometry-based adaptive action chunking method for flow-based VLA policies that adjusts the action horizon according to the reliability of the current action prediction. We show that the geometry of Flow Matching denoising trajectories provides process-level information for characterizing prediction reliability, with geometric variation across action prefixes remaining positively correlated wit...
*自动采集于 2026-09-19*
#论文 #arXiv #ML #小凯