Adaptive action chunking improves robot task control accuracy
GeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Trajectories in VLA Policies
RoboticsArtificial IntelligenceMachine Learning
Summary
Robots performing tasks often follow sequences of actions, but the best length of these action chunks can vary depending on the stage of the task. The authors propose a method called GeoAAC that adjusts the length of these action chunks dynamically, based on how reliable the robot’s current action predictions are. They use geometric insights from the robot’s decision-making process to decide when to change the chunk size, improving task success rates without needing extra training. Tests on various simulated and real-world robot tasks show this method helps robots perform better compared to fixed or other adaptive strategies.
What this means in practice
- •For robotics engineers: Improve robot control strategies by dynamically adjusting action sequence lengths based on prediction reliability to increase precision and success rates in task execution.
- •For industrial automation teams: Enhance performance and reliability of automated assembly or manipulation systems using adaptable action chunking rather than fixed action intervals.
Authors
Xin Chen, Sen Chen, Yujuan Ding, Jian Liu, Guoqing Wang, Wei Ye, Heng Tao Shen, Yi Bin
Abstract
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 \textbf{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 with predictive uncertainty. GeoAAC uses this prefix-wise geometry to construct a horizon-wise geometric profile and adaptively determine the action horizon from a single generation without additional training. Experiments with GR00T N1.5 and π0.5 on LIBERO, LIBERO-Pro, RoboCasa365, and real-world manipulation tasks show consistent improvements over fixed-action-horizon baselines and existing adaptive methods, including up to 8.7 percentage points in simulation and an increase in average real-world success rate from 53.3\% to 74.4\%.