# ACG算法免训练提升VLA模型动作连贯性与操作成功率

> 原文链接：https://robotics.hellobike.com/papers/2025-10-25-2510-22201-c1b3bae4
> 频道：papers | 更新日期：2025-10-25 | 标签：auto, arxiv, cs.RO
> 核心摘要：针对流匹配VLA模型模仿学习中的演示噪声问题，提出免训练测试时引导算法ACG。在RoboCasa、DexMimicGen及SO-101真机任务中验证，该方案有效抑制轨迹漂移并显著提升精细操作成功率，代码已开源。

# ACG算法免训练提升VLA模型动作连贯性与操作成功率

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[ACG: Action Coherence Guidance for Flow-based Vision-Language-Action models](https://arxiv.org/abs/2510.22201)（arXiv 2510.22201）

## 核心内容

扩散与流匹配模型虽增强了VLA泛化能力，但高生成容量使其对模仿学习数据中的抖动、暂停等噪声极度敏感，导致部署时动作连贯性下降与轨迹漂移，严重影响精细操作。本文提出Action Coherence Guidance (ACG) 算法，作为一种无需重训练的测试时引导机制，专门解决VLA模型在复杂指令下的动作不一致问题，并在仿真与真机环境中验证了其有效性。

ACG通过在推理阶段施加连贯性约束来修正生成策略，避免了高昂的重新训练成本。研究团队在RoboCasa、DexMimicGen基准及SO-101机器人真机任务上进行评估，结果显示该方法一致性地改善了动作平滑度并提升了任务成功率。相关代码与项目主页已在GitHub开源，为VLA模型的落地部署提供了低成本的稳定性增强方案。

## 关键看点

- **核心痛点**：解决流匹配VLA模型因人类演示噪声导致的动作抖动与轨迹漂移，防止精细操作失败
- **技术机制**：提出免训练测试时引导算法ACG，在推理阶段直接增强动作连贯性，无需修改模型权重
- **实验验证**：在RoboCasa、DexMimicGen仿真及SO-101真机任务中均实现成功率提升与动作优化
- **开源资产**：完整代码与项目文档已在GitHub开源，支持社区复现与二次开发

## 工程与产业观察

ACG为VLA模型从仿真走向真机提供了关键的工程补丁。在不增加训练算力前提下，通过测试时引导解决模仿学习固有的噪声敏感问题，显著降低了精细操作任务的部署门槛。这种免训练后处理范式对提升具身智能体在非结构化环境中的鲁棒性具有重要参考价值。

## 英文摘要 (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.

作者：Minho Park, Kinam Kim, Junha Hyung, Hyojin Jang, Hoiyeong Jin, Jooyeol Yun, Hojoon Lee, Jaegul Choo
:::

## 原始参考来源 (Sources)

- [arXiv 2510.22201：ACG: Action Coherence Guidance for Flow-based Vision-Language-Action models](https://arxiv.org/abs/2510.22201)