# VLA模型LIBERO基准91.7%成功率可由kNN检索实现

> 原文链接：https://robotics.hellobike.com/papers/2026-11-11-2611-13160-d97648b0
> 频道：papers | 更新日期：2026-11-11 | 标签：auto, arxiv
> 核心摘要：研究揭示VLA模型在LIBERO仿真中95%成功率主要依赖训练数据检索而非泛化。无策略训练的kNN检索达91.7%成功率且抗扰最优，提出LAC指标量化表征空间动作一致性并开源代码。

# VLA模型LIBERO基准91.7%成功率可由kNN检索实现

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[How Much of Your VLA Is kNN](https://arxiv.org/abs/2611.13160)（arXiv 2611.13160）

## 核心内容

针对VLA模型在LIBERO基准超95%成功率的真实性能来源，研究表明仅通过检索邻近演示并重用动作块，无需策略训练即可达91.7%成功率。在LIBERO-Plus扰动测试中，该检索方法性能衰减幅度为所有评估策略中最小。对九款公开VLA的分析证实，微调主要优化了视觉语言骨干作为检索键的能力，模型输出动作仍高度贴近原始演示数据。

研究提出Local Action Consistency（LAC）离线指标，用于度量表征空间中邻近观测是否关联相似未来动作，实验验证LAC与检索成功率强相关并可指导检索键训练。这表明当前VLA可能更多充当高效记忆检索器而非通用策略学习器。相关评估工具与源码已在 `lerobot_knn` 仓库开源，为具身智能算法从记忆机制向真泛化演进提供量化基准。

## 关键看点

- **kNN检索基线**：无策略训练下重用演示动作块在LIBERO达91.7%成功率，逼近VLA模型95%上限
- **抗扰动鲁棒性**：在LIBERO-Plus环境扰动测试中，纯检索策略的性能下降幅度低于所有已评估VLA
- **LAC离线指标**：量化表征空间邻近观测的动作一致性，可有效追踪检索成功率并指导检索键优化
- **VLA机制归因**：微调主要提升视觉语言骨干检索键质量，模型动作输出未显著脱离演示数据分布

## 工程与产业观察

该研究警示当前VLA高成功率或源于数据记忆而非物理泛化，LAC指标与kNN基线为区分记忆与推理提供工程标尺。开发者应警惕仿真过拟合，在真机部署前引入检索基线对照与表征一致性校验，避免将高效索引误判为具身智能涌现。

## 英文摘要 (Abstract)

Vision-language-action (VLA) models now exceed 95% success on the LIBERO simulation benchmark, yet it remains unclear how much of this performance requires generalization beyond the training demonstrations. In this work, we show that retrieving nearby demonstrations and reusing their action chunks achieves a success rate of 91.7 % on LIBERO without policy training. Under LIBERO-Plus perturbations, retrieval shows the smallest drop of all evaluated policies. Across nine publicly available VLAs, fine-tuning mostly makes the vision-language backbone a better retrieval key, and the actions of the VLAs remain close to the demonstrations. In addition, we introduce Local Action Consistency (LAC), an offline metric that measures whether nearby observations in representation space are also associated with similar future actions. We show that LAC tracks the success rate of retrieval and can guide the training of a retrieval key. Our evaluation and source code are available at https://github.com/sereact/lerobot_knn.

作者：Denis Grachev, Damien LaRocque, Omar Moured
:::

## 原始参考来源 (Sources)

- [arXiv 2611.13160：How Much of Your VLA Is kNN](https://arxiv.org/abs/2611.13160)