ACG: Action Coherence Guidance for Flow-based VLA models

2025-10-25

AI Summary

Action noise—such as jitter and pauses—in human demonstrations degrades trajectory coherence in flow-matching-based vision-language-action (VLA) models, leading to deployment instability and failure in fine-grained manipulation. To address this, we propose a **training-free, test-time action coherence guidance method** that dynamically refines action sequences during inference to enhance smoothness and temporal consistency, significantly improving robustness to demonstration noise. Our approach is framework-agnostic, seamlessly integrating with both diffusion and flow-matching VLA architectures without introducing additional parameters or training overhead. We evaluate it on RoboCasa, DexMimicGen, and real-world SO-101 tasks, demonstrating substantial improvements in action coherence metrics and task success rates. The method provides a lightweight, general-purpose, plug-and-play stability enhancement for practical VLA deployment.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Diffusion and flow matching models have emerged as powerful robot policies, enabling Vision-Language-Action (VLA) models to generalize across diverse scenes and instructions. Yet, when trained via imitation learning, their high generative capacity makes them sensitive to noise in human demonstrations: jerks, pauses, and jitter which reduce action coherence. Reduced action coherence causes instability and trajectory drift during deployment, failures that are catastrophic in fine-grained manipulation where precision is crucial. In this paper, we present Action Coherence Guidance (ACG) for VLA models, a training-free test-time guidance algorithm that improves action coherence and thereby yields performance gains. Evaluated on RoboCasa, DexMimicGen, and real-world SO-101 tasks, ACG consistently improves action coherence and boosts success rates across diverse manipulation tasks. Code and project page are available at https://github.com/DAVIAN-Robotics/ACG and https://DAVIAN-Robotics.github.io/ACG , respectively.

Problem

• Improving action coherence in vision-language-action models
• Reducing trajectory drift during robotic manipulation deployment
• Mitigating noise sensitivity in imitation learning policies

Innovation

• Training-free test-time guidance algorithm for VLA models
• Improves action coherence in vision-language-action policies
• Reduces trajectory drift and boosts manipulation success rates