# SN-TRV实现VLA跨任务防御迁移，PiPER抗攻击成功率升至45.8%

> 原文链接：https://robotics.hellobike.com/papers/2026-11-17-2611-18702-2e0d6454
> 频道：papers | 更新日期：2026-11-17 | 标签：auto, arxiv
> 核心摘要：提出SN-TRV方法将源任务鲁棒性零样本迁移至目标VLA策略，无需目标对抗训练。在松灵PiPER真机测试中，面对打印对抗补丁的平均操作成功率从20.3%提升至45.8%，验证了视觉编码器防御能力的跨任务复用性。

# SN-TRV实现VLA跨任务防御迁移，PiPER抗攻击成功率升至45.8%

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[Robustness as a Transferable Skill: Cross-Task Adversarial Defense for Vision-Language-Action Policies](https://arxiv.org/abs/2611.18702)（arXiv 2611.18702）

## 核心内容

针对VLA策略易受视觉对抗扰动且传统防御依赖特定任务微调的痛点，研究提出Source-Norm Transferable Robustness Vector（SN-TRV）方法。该方法证实仅凭源任务防御更新即可提升未见任务的鲁棒性，并在三种VLA骨干网络及固定/自适应攻击下，性能均优于朴素平均法，确立了跨任务防御复用的可行性。

SN-TRV通过保留源更新均值方向并利用源更新范数校准幅度，直接将防御向量应用于目标视觉编码器，全程无需目标对抗数据、梯度优化或额外推理模块。在松灵PiPER机械臂真机实验中，面对打印对抗补丁干扰，该方法的平均任务成功率由基线的20.3%显著提升至45.8%，有效解决了具身智能落地中的长尾安全难题。

## 关键看点

- **SN-TRV机制**：保留源更新均值方向并用范数校准幅度，实现防御向量向目标视觉编码器的零样本迁移
- **真机性能验证**：松灵PiPER在打印对抗补丁干扰下，平均操作成功率从20.3%提升至45.8%
- **零成本部署**：无需目标侧对抗数据、梯度优化或新增推理模块，单次源防御更新即生效
- **泛化能力确认**：在三种VLA骨干及多种攻击模式下验证有效，证明鲁棒性可作为独立技能跨任务迁移

## 工程与产业观察

该研究将VLA鲁棒性解耦为可迁移技能，SN-TRV允许在不重训目标模型的前提下继承源任务防御能力。这对解决具身智能多场景部署中的数据稀缺与安全合规问题具有工程价值，提示行业应重视构建通用视觉防御基座而非单点微调。

## 英文摘要 (Abstract)

Vision-Language-Action (VLA) policies are vulnerable to adversarial visual perturbations, while defenses often require task-specific adversarial fine-tuning. We investigate whether robustness learned on source manipulation tasks can protect independently specialized target policies without target-side adversarial training. We first show that even a single source defense update improves robustness on held-out tasks, establishing cross-task defense reuse independently of multi-source aggregation. To combine source updates, we introduce Source-Norm Transferable Robustness Vector (SN-TRV), which retains the direction of their mean while calibrating its magnitude using source-update norms. The resulting update is applied to the target visual encoder without target adversarial data, target-side gradient optimization, or additional inference modules. Evaluations across three VLA backbones, fixed and adaptive attacks, and different source defense recipes show improved robustness over naive averaging and identify limits in attack coverage. Physical PiPER trials increase average success under printed adversarial patches from 20.3% to 45.8%.

作者：Jinquan Zhang, Run Yang, Dongfu Yin, Qiang Lai, Yufeng Yan, Pengshuai Yin, F. Richard Yu
:::

## 原始参考来源 (Sources)

- [arXiv 2611.18702：Robustness as a Transferable Skill: Cross-Task Adversarial Defense for Vision-Language-Action Policies](https://arxiv.org/abs/2611.18702)