LiQiiiii/Awesome-VLA-Safety
GitHub: LiQiiiii/Awesome-VLA-Safety
围绕视觉-语言-动作模型安全性的综述资源库，系统整理了攻击威胁、防御机制与评估基准方面的前沿研究论文和工具。
Stars: 122 | Forks: 3
视觉-语言-动作安全性：威胁、挑战、评估与机制
关于视觉-语言-动作模型安全性的统一且最新的概述
[](https://github.com/LiQiiiii/Awesome-VLA-Safety/pulls)
Figure 1. An overview of the VLA safety landscape covered in this survey.
Figure 2. Training-time attacks on VLA models.
Figure 3. Taxonomy of inference-time safety and robustness in VLA models.
[DropVLA：一种针对视觉-语言-动作模型的动作级后门攻击](https://arxiv.org/pdf/2510.10932)
Xu, Zonghuan and Li, Jiayu and Zhao, Yunhan and Zheng, Xiang and Ma, Xingjun and Jiang, Yu-Gang |
|[Github](https://github.com/megaknight114/DropVLA) [论文](https://arxiv.org/pdf/2510.10932)|[//]: #04/19 |[](https://github.com/Zxy-MLlab/BadVLA)
[BadVLA：通过目标解耦优化对视觉-语言-动作模型进行后门攻击](https://arxiv.org/pdf/2505.16640)
Zhou, Xueyang and Tie, Guiyao and Zhang, Guowen and Wang, Hechang and Zhou, Pan and Sun, Lichao |
|[Github](https://github.com/Zxy-MLlab/BadVLA) [论文](https://arxiv.org/pdf/2505.16640)|[//]: #04/19 |[通过物理对象对视觉-语言-动作模型进行目标导向的后门攻击](https://arxiv.org/pdf/2510.09269)
Zhou, Zirun and Xiao, Zhengyang and Xu, Haochuan and Sun, Jing and Wang, Di and Zhang, Jingfeng |
|[Github](https://goba-attack.github.io/) [论文](https://arxiv.org/pdf/2510.09269)|[//]: #04/19 #### 时间与状态空间后门 [🔝 返回顶部](#quick-links) | 标题与作者 | 简介 | 链接 | |:--| :----: | :---:| |[State Backdoor：在状态空间中对视觉-语言-动作模型进行隐蔽的真实世界投毒攻击](https://arxiv.org/pdf/2601.04266)
Guo, Ji and Jiang, Wenbo and Lin, Yansong and Liu, Yijing and Zhang, Ruichen and Lu, Guomin and Chen, Aiguo and Han, Xinshuo and Li, Hongwei and Niyato, Dusit |
|[论文](https://arxiv.org/pdf/2601.04266)|[//]: #04/19
|[SilentDrift：利用动作分块对视觉-语言-动作模型进行隐蔽后门攻击](https://arxiv.org/pdf/2601.14323) Xu, Bingxin and Shang, Yuzhang and Wang, Binghui and Ferrara, Emilio |
|[论文](https://arxiv.org/pdf/2601.14323)|[//]: #04/19
### 训练阶段防御
#### 以数据、感知和奖励为中心的对齐
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[](https://github.com/luffycodes/Tutorbot-Spock)[大型语言模型的教学对齐](https://arxiv.org/pdf/2402.05000)
Sonkar, Shashank and Ni, Kangqi and Chaudhary, Sapana and Baraniuk, Richard G. |
|[Github](https://github.com/luffycodes/Tutorbot-Spock) [论文](https://arxiv.org/pdf/2402.05000)|[//]: #04/19 |[](https://github.com/AIGeeksGroup/EvoVLA)
[EvoVLA：自进化的视觉-语言-动作模型](https://arxiv.org/pdf/2511.16166)
Liu, Zeting and Yang, Zida and Zhang, Zeyu and Tang, Hao |
|[Github](https://github.com/AIGeeksGroup/EvoVLA) [论文](https://arxiv.org/pdf/2511.16166)|[//]: #04/19 |[用于端到端自动驾驶的生成式场景推演 (GeRo)](https://arxiv.org/pdf/2601.11475)
Yasarla, Rajeev and Hegde, Deepti and Han, Shizhong and Cheng, Hsin-Pai and Shi, Yunxiao and Sadeghigooghari, Meysam and Mahajan, Shweta and Bhattacharyya, Apratim and Liu, Litian and Garrepalli, Risheek and Svantesson, Thomas and Porikli, Fatih and Cai, Hong |
|[论文](https://arxiv.org/pdf/2601.11475)|[//]: #04/19
|[Safe-Night VLA：通过面向安全关键操作的感知热量的视觉-语言-动作模型看见未见之物](https://arxiv.org/pdf/2603.05754) Yu, Dian and Zhou, Qingchuan and Huang, Bingkun and Khadiv, Majid and Yang, Zewen |
|[论文](https://arxiv.org/pdf/2603.05754)|[//]: #04/19
#### 以策略为中心的安全优化
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[](https://github.com/PKU-Alignment/SafeVLA)[SafeVLA：通过约束学习实现视觉-语言-动作模型的安全对齐](https://arxiv.org/pdf/2503.03480)
Zhang, Borong and Zhang, Yuhao and Ji, Jiaming and Lei, Yingshan and Dai, Josef and Chen, Yuanpei and Yang, Yaodong |
|[Github](https://github.com/PKU-Alignment/SafeVLA) [论文](https://arxiv.org/pdf/2503.03480)|[//]: #04/19 |[用于机器人操作的人机协同在线拒绝采样 (Hi-ORS)](https://arxiv.org/pdf/2510.26406)
Lu, Guanxing and Zhao, Rui and Lin, Haitao and Zhang, He and Tang, Yansong |
|[Github](https://hiors-project.github.io/) [论文](https://arxiv.org/pdf/2510.26406)|[//]: #04/19 |[通过多目标策略优化进行安全优化的强化学习 (SORL)](https://arxiv.org/pdf/2402.15197)
Honari, Homayoun and Tamizi, Mehran Ghafarian and Najjaran, Homayoun |
|[论文](https://arxiv.org/pdf/2402.15197)|[//]: #04/19
|[VLA-Forget：面向具身基础模型的视觉-语言-动作遗忘](https://arxiv.org/pdf/2604.03956) Ranjan, Ravi and Polyzou, Agoritsa |
|[论文](https://arxiv.org/pdf/2604.03956)|[//]: #04/25
#### 人机协同策略优化
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[](https://github.com/GeWu-Lab/Action-Preference-Optimization)[通过动作偏好优化 (APO) 实现人类辅助的机器人策略优化](https://arxiv.org/pdf/2506.07127)
Xia, Wenke and Yang, Yichu and Wu, Hongtao and Ma, Xiao and Kong, Tao and Hu, Di |
|[Github](https://github.com/GeWu-Lab/Action-Preference-Optimization) [论文](https://arxiv.org/pdf/2506.07127)|[//]: #04/19 ### 推理阶段攻击 #### 语义越狱 [🔝 返回顶部](#quick-links) | 标题与作者 | 简介 | 链接 | |:--| :----: | :---:| |[越狱 LLM 控制的机器人](https://arxiv.org/pdf/2410.13691)
Robey, Alexander and Ravichandran, Zachary and Kumar, Vijay and Hassani, Hamed and Pappas, George J |
|[论文](https://arxiv.org/pdf/2410.13691)|[//]: #04/19
|[](https://github.com/embodied-llms-safety.github.io/)[Badrobot：在物理世界中越狱具身 LLM](https://arxiv.org/pdf/2407.20242)
Zhang, Hangtao and Zhu, Chenyu and Wang, Xianlong and Zhou, Ziqi and, Changgan and Li, Minghui and Xue, Lulu and Wang, Yichen and Hu, Shengshan and Liu, Aishan and others |
|[Github](https://embodied-llms-safety.github.io/) [论文](https://arxiv.org/pdf/2407.20242)|[//]: #04/19 |[](https://github.com/eliotjones1/robogcg)
[针对机器人视觉-语言-动作模型的对抗攻击](https://arxiv.org/pdf/2506.03350)
Jones, Eliot Krzysztof and Robey, Alexander and Zou, Andy and Ravichandran, Zachary and Pappas, George J and Hassani, Hamed and Fredrikson, Matt and Kolter, J Zico |
|[Github](https://github.com/eliotjones1/robogcg) [论文](https://arxiv.org/pdf/2506.03350)|[//]: #04/19 |[](https://github.com/verazuo/jailbreak_llms)
["现在做任何事"：在大型语言模型上表征和评估野外的越狱提示词](https://arxiv.org/pdf/2308.03825)
Shen, Xinyue and Chen, Zeyuan and Backes, Michael and Shen, Yun and Zhang, Yang |
|[Github](https://github.com/verazuo/jailbreak_llms) [论文](https://arxiv.org/pdf/2308.03825)|[//]: #04/19 #### 视觉扰动 [🔝 返回顶部](#quick-links) | 标题与作者 | 简介 | 链接 | |:--| :----: | :---:| |[](https://github.com/William-wAng618/roboticAttack)
[探索机器人技术中视觉-语言-动作模型的对抗脆弱性](https://arxiv.org/pdf/2411.13587)
Wang, Taowen and Han, Cheng and Liang, James and Yang, Wenhao and Liu, Dongfang and Zhang, Luna Xinyu and Wang, Qifan and Luo, Jiebo and Tang, Ruixiang |
|[Github](https://github.com/William-wAng618/roboticAttack) [论文](https://arxiv.org/pdf/2411.13587)|[//]: #04/19 |[当对齐失败时：针对视觉-语言-动作模型的多模态对抗攻击](https://arxiv.org/abs/2511.16203)
Yan, Yuping and Xie, Yuhan and Zhang, Yixin and Lyu, Lingjuan and Wang, Handing and Jin, Yaochu |
|[论文](https://arxiv.org/abs/2511.16203)|[//]: #04/19
|[](https://github.com/xinwong/FreezeVLA)[FreezeVLA：针对视觉-语言-动作模型的动作冻结攻击](https://arxiv.org/abs/2509.19870)
Wang, Xin and Li, Jie and Weng, Zejia and Wang, Yixu and Gao, Yifeng and Pang, Tianyu and Du, Chao and Teng, Yan and Wang, Yingchun and Wu, Zuxuan and others |
|[Github](https://github.com/xinwong/FreezeVLA) [论文](https://arxiv.org/abs/2509.19870)|[//]: #04/19 |[如果你在等待信号……那可能并不是它！缓解视觉-语言智能体系统因视觉注入引起的信任边界混淆](https://arxiv.org/abs/2604.19844v1)
Jiamin Chang and Minhui Xue and Ruoxi Sun and Shuchao Pang and Salil S. Kanhere and Hammond Pearce |
|[论文](https://arxiv.org/abs/2604.19844v1)|[//]: #04/25
#### 物理干预
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[幻影威胁：探索并增强 VLA 模型抵御物理传感器攻击的鲁棒性](https://arxiv.org/abs/2511.10008) Lu, Xuancun and Chen, Jiaxiang and Xiao, Shilin and Jin, Zizhi and Chen, Zhangrui and Yu, Hanwen and Qian, Bohan and Zhou, Ruochen and Ji, Xiaoyu and Xu, Wenyuan |
|[论文](https://arxiv.org/abs/2511.10008)|[//]: #04/19
|[State Backdoor：在状态空间中对视觉-语言-动作模型进行隐蔽的真实世界投毒攻击](https://arxiv.org/abs/2601.04266) Ji Guo, Wenbo Jiang, Yansong Lin, Yijing Liu, Ruichen Zhang, Guomin Lu, Aiguo Chen, Xinshuo Han, Hongwei Li, Dusit Niyato |
|[论文](https://arxiv.org/abs/2601.04266)|[//]: #04/19
### 推理阶段防御
#### 决策层防护栏
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[](https://github.com/THU-RCSCT/vlsa-aegis)[VLSA：带有即插即用安全约束层的视觉-语言-动作模型](https://arxiv.org/abs/2512.11891)
Hu, Songqiao and Liu, Zeyi and Liu, Shuang and Cen, Jun and Meng, Zihan and He, Xiao |
|[Github](https://github.com/THU-RCSCT/vlsa-aegis) [论文](https://arxiv.org/abs/2512.11891)|[//]: #04/19 |[用于 LLM 驱动机器人的安全防护栏](https://arxiv.org/pdf/2503.07885)
Zachary Ravichandran, Alexander Robey, Vijay Kumar, George J. Pappas, Hamed Hassani |
|[论文](https://arxiv.org/pdf/2503.07885)|[//]: #04/19
|[从文字到安全：用于机器人导航的语言条件安全过滤](https://arxiv.org/abs/2511.05889) Zeyuan Feng, Haimingyue Zhang, Somil Bansal |
|[论文](https://arxiv.org/abs/2511.05889)|[//]: #04/20
#### 运行时监控
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[运行时观测干预使视觉-语言-动作模型在视觉上更鲁棒](https://arxiv.org/abs/2410.01971) Asher J. Hancock, Allen Z. Ren, Anirudha Majumdar |
|[论文](https://arxiv.org/abs/2410.01971)|[//]: #04/20
|[Safe-vln：在连续环境中运行的自主机器人视觉与语言导航的避障研究](https://arxiv.org/pdf/2311.02817) Lu Yue, Dongliang Zhou, Liang Xie, Feitian Zhang, Ye Yan, Erwei Yin |
|[论文](https://arxiv.org/pdf/2311.02817)|[//]: #04/20
|[可供性场干预：使 VLA 在机器人操作中逃离记忆陷阱](https://arxiv.org/pdf/2512.07472) Siyu Xu, Zijian Wang, Yunke Wang, Chenghao Xia, Tao Huang, Chang Xu |
|[论文](https://arxiv.org/pdf/2512.07472)|[//]: #04/20
|[](https://github.com/real-stanford/reflect)[Reflect：总结机器人经验以进行失败解释与纠正](https://arxiv.org/abs/2306.15724)
Zeyi Liu, Arpit Bahety, Shuran Song |
|[Github](https://github.com/real-stanford/reflect) [论文](https://arxiv.org/abs/2306.15724)|[//]: #04/20 |[Failsafe：视觉-语言-动作模型中失败的推理与恢复](https://arxiv.org/abs/2510.01642)
Zijun Lin, Jiafei Duan, Haoquan Fang, Dieter Fox, Ranjay Krishna, Cheston Tan, Bihan Wen |
|[论文](https://arxiv.org/abs/2510.01642)|[//]: #04/20
|[具备运行时安全监督的视觉-语言-动作驾驶的因果场景叙述](https://arxiv.org/abs/2604.01723v1) Li, Yun and Zhang, Yidu and Thompson, Simon and Javanmardi, Ehsan and Tsukada, Manabu |
|[论文](https://arxiv.org/abs/2604.01723v1)|[//]: #04/25
#### 物理故障保护
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[利用归纳偏置构建安全的机器人基础模型](https://arxiv.org/abs/2505.10219) Maximilian Tölle, Theo Gruner, Daniel Palenicek, Tim Schneider, Jonas Günster, Joe Watson, Davide Tateo, Puze Liu, Jan Peters |
|[论文](https://arxiv.org/abs/2505.10219)|[//]: #04/20
|[DroneVLA：基于 VLA 的空中操作](https://arxiv.org/abs/2601.13809) Fawad Mehboob, Monijesu James, Amir Habel, Jeffrin Sam, Miguel Altamirano Cabrera, Dzmitry Tsetserukou |
|[论文](https://arxiv.org/abs/2601.13809)|[//]: #04/20
|[CompliantVLA-adaptor：用于安全富接触操作的 VLM 引导变阻抗动作](https://arxiv.org/abs/2601.15541) Heng Zhang, Wei-Hsing Huang, Qiyi Tong, Gokhan Solak, Puze Liu, Kaidi Zhang, Sheng Liu, Jan Peters, Yu She, Arash Ajoudani |
|[论文](https://arxiv.org/abs/2601.15541)|[//]: #04/20
### 推理阶段评估与基准
#### 多层级基准
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[VLA-Arena：一个用于基准测试视觉-语言-动作模型的开源框架](https://arxiv.org/abs/2512.22539) Borong Zhang, Jiahao Li, Jiachen Shen, Yuai Cai, Yuhao Zhang, Yuanpei Chen, Juntao Dai, Jiaming Ji, Yaodong Yang |
|[论文](https://arxiv.org/abs/2512.22539)|[//]: #04/20
|[生成机器人准则及语义安全基准](https://arxiv.org/abs/2503.08663) Pierre Sermanet, Anirudha Majumdar, Alex Irpan, Dmitry Kalashnikov, Vikas Sindhwani |
|[论文](https://arxiv.org/abs/2503.08663)|[//]: #04/20
|[视觉-语言模型能理解建筑工人吗？一项探索性研究](https://arxiv.org/abs/2601.10835) Hieu Bui, Nathaniel E. Chodosh, Arash Tavakoli |
|[论文](https://arxiv.org/abs/2601.10835)|[//]: #04/20
|[VLA 在开放世界环境中（实际上）是如何工作的](https://arxiv.org/abs/2604.21192v1) Amir Rasouli, Yangzheng Wu, Zhiyuan Li, Rui Heng Yang, Xuan Zhao, Charles Eret, Sajjad Pakdamansavoji |
|[论文](https://arxiv.org/abs/2604.21192v1)|[//]: #04/25
|[HazardArena：评估视觉-语言-动作模型中的语义安全性](https://arxiv.org/abs/2604.12447v1) Chen, Zixing and Gao, Yifeng and Wang, Li and Zhao, Yunhan and Liu, Yi and Li, Jiayu and Zheng, Xiang and Wu, Zuxuan and Wang, Cong and Ma, Xingjun and others |
|[论文](https://arxiv.org/abs/2604.12447v1)|[//]: #04/25
|[](https://github.com/icrdrive/icrdrive)[ICR-Drive：用于端到端语言驱动自动驾驶的指令反事实鲁棒性](https://arxiv.org/abs/2604.05378v1)
Hamid, Kaiser and Cui, Can and Liang, Nade |
|[Github](https://github.com/icrdrive/icrdrive) [论文](https://arxiv.org/abs/2604.05378v1)|[//]: #04/25 #### 不确定性校准 [🔝 返回顶部](#quick-links) | 标题与作者 | 简介 | 链接 | |:--| :----: | :---:| |[Vlatest：测试和评估用于机器人操作的视觉-语言-动作模型](https://arxiv.org/abs/2409.12894)
Zhijie Wang, Zhehua Zhou, Jiayang Song, Yuheng Huang, Zhan Shu, Lei Ma |
|[论文](https://arxiv.org/abs/2409.12894)|[//]: #04/20
|[ROVER：基于调节器的黑盒机器人策略鲁棒性时间验证](https://arxiv.org/abs/2511.17781) Kristy Sakano, Jianyu An, Dinesh Manocha, Huan Xu |
|[论文](https://arxiv.org/abs/2511.17781)|[//]: #04/20
|[](https://github.com/vla-safe/SAFE)[SAFE：用于视觉-语言-动作模型的多任务失败检测](https://arxiv.org/abs/2506.09937)
Qiao Gu, Yuanliang Ju, Shengxiang Sun, Igor Gilitschenski, Haruki Nishimura, Masha Itkina, Florian Shkurti |
|[Github](https://github.com/vla-safe/SAFE) [论文](https://arxiv.org/abs/2506.09937)|[//]: #04/20 |[视觉-语言-动作模型的置信度校准](https://arxiv.org/abs/2507.17383)
Thomas P Zollo, Richard Zemel |
|[论文](https://arxiv.org/abs/2507.17383)|[//]: #04/20
|[评估视觉-语言-动作驱动机器人的不确定性与质量](https://arxiv.org/abs/2507.17049) Pablo Valle, Chengjie Lu, Shaukat Ali, Aitor Arrieta |
|[论文](https://arxiv.org/abs/2507.17049)|[//]: #04/20
### 评估
**安全基准**
#### 对抗鲁棒性基准
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[VLA-Risk：以物理鲁棒性对视觉-语言-动作模型进行基准测试](https://openreview.net/forum?id=31EjDFwFEe) Ru, Yanchi and Zhao, Zhengyue and Ma, Yingzi and Liu, Xiaogeng and Xiao, Chaowei |— |[论文](https://openreview.net/forum?id=31EjDFwFEe)|[//]: #04/21 |[VLATest：用于机器人操作的视觉-语言-动作模型的测试与评估](https://arxiv.org/abs/2409.12894)
Wang, Zhijie and Zhou, Zhehua and Song, Jiayang and Huang, Yuheng and Shu, Zhan and Ma, Lei |
|[论文](https://arxiv.org/abs/2409.12894)|[//]: #04/21
|[探索机器人技术中视觉-语言-动作模型的对抗脆弱性](https://arxiv.org/abs/2411.13587) Wang, Taowen and Han, Cheng and Liang, James and Yang, Wenhao and Liu, Dongfang and Zhang, Luna Xinyu and Wang, Qifan and Luo, Jiebo and Tang, Ruixiang |
|[项目](https://vlaattacker.github.io/) [论文](https://arxiv.org/abs/2411.13587)|[//]: #04/21 |[](https://github.com/ZJUshine/lamps-vla-robustness.github.io)
[探索视觉-语言-动作模型对传感器攻击的鲁棒性](https://dl.acm.org/doi/10.1145/3733800.3763262)
Lu, Xuancun and Chen, Jiaxiang and Xiao, Shilin and Jin, Zizhi and Zhou, Ruochen and Ji, Xiaoyu and Xu, Wenyuan |
|[Github](https://github.com/ZJUshine/lamps-vla-robustness.github.io) [论文](https://dl.acm.org/doi/pdf/10.1145/3733800.3763262)|[//]: #04/21 #### 任务级安全基准 [🔝 返回顶部](#quick-links) | 标题与作者 | 简介 | 链接 | |:--| :----: | :---:| |[](https://github.com/shengyin1224/SafeAgentBench)
[SafeAgentBench：针对具身 LLM 智能体安全任务规划的基准](https://arxiv.org/abs/2412.13178)
Yin, Sheng and Pang, Xianghe and Ding, Yuanzhuo and Chen, Menglan and Bi, Yutong and Xiong, Yichen and Huang, Wenhao and Xiang, Zhen and Shao, Jing and Chen, Siheng |
|[Github](https://github.com/shengyin1224/SafeAgentBench) [论文](https://arxiv.org/abs/2412.13178)|[//]: #04/21 |[AgentSafe：针对具身智能体危险指令的安全性基准测试](https://arxiv.org/abs/2506.14697)
Ying, Zonghao and Wang, Le and Xiao, Yisong and Wang, Jiakai and Ma, Yuqing and Guo, Jinyang and Yin, Zhenfei and Zhang, Mingchuan and Liu, Aishan and Liu, Xianglong |
|[论文](https://arxiv.org/abs/2506.14697)|[//]: #04/21
|[SafeMind：具身 LLM 智能体安全风险的基准测试与缓解](https://arxiv.org/abs/2509.25885) Chen, Ruolin and Sun, Yinqian and Wang, Jihang and Lv, Mingyang and Zhang, Qian and Zeng, Yi |
|[论文](https://arxiv.org/abs/2509.25885)|[//]: #04/21
#### 综合能力与安全基准
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[](https://github.com/PKU-Alignment/VLA-Arena)[VLA-Arena：一个用于基准测试视觉-语言-动作模型的开源框架](https://arxiv.org/abs/2512.22539)
Zhang, Borong and Li, Jiahao and Shen, Jiachen and Cai, Yishuai and Zhang, Yuhao and Chen, Yuanpei and Dai, Juntao and Ji, Jiaming and Yang, Yaodong |
|[项目](https://vla-arena.github.io/) [Github](https://github.com/PKU-Alignment/VLA-Arena)
[论文](https://arxiv.org/abs/2512.22539)|[//]: #04/21 |[](https://github.com/OpenMOSS/VLABench)
[VLABench：一个面向具有长程推理任务的语言条件机器人操作的大规模基准](https://arxiv.org/abs/2412.18194)
Zhang, Shiduo and Xu, Zhe and Liu, Peiju and Yu, Xiaopeng and Li, Yuan and Gao, Qinghui and Fei, Zhaoye and Yin, Zhangyue and Wu, Zuxuan and Jiang, Yu-Gang and Qiu, Xipeng |
|[Github](https://github.com/OpenMOSS/VLABench) [论文](https://arxiv.org/abs/2412.18194)|[//]: #04/21 |[](https://github.com/Lifelong-Robot-Learning/LIBERO)
[LIBERO：面向终身机器人学习的知识迁移基准测试](https://arxiv.org/abs/2306.03310)
Liu, Bo and Zhu, Yifeng and Gao, Chongkai and Feng, Yihao and Liu, Qiang and Zhu, Yuke and Stone, Peter |
|[Github](https://github.com/Lifelong-Robot-Learning/LIBERO) [论文](https://arxiv.org/abs/2306.03310)|[//]: #04/21 |[](https://github.com/Zxy-MLlab/LIBERO-PRO)
[LIBERO-PRO：迈向超越记忆的视觉-语言-动作模型鲁棒与公平评估](https://arxiv.org/abs/2510.03827)
Zhou, Xueyang and Xu, Yangming and Tie, Guiyao and Chen, Yongchao and Zhang, Guowen and Chu, Duanfeng and Zhou, Pan and Sun, Lichao |
|[Github](https://github.com/Zxy-MLlab/LIBERO-PRO) [论文](https://arxiv.org/abs/2510.03827)|[//]: #04/21 |[CostNav：一个用于具身智能体成本感知评估的导航基准](https://arxiv.org/abs/2511.20216)
Seong, Haebin and Kim, Sungmin and Kim, Minchan and Cho, Yongjun and Joe, Myunchul and Choi, Suhwan and Jung, Jaeyoon and Youn, Jiyong and Kim, Yoonshik and Seong, Samwoo and others |
|[论文](https://arxiv.org/abs/2511.20216)|[//]: #04/21
#### 越狱与对齐基准
[🔝 返回顶部](#quick-links)
| 标题与作者 | 简介 | 链接 |
|:--| :----: | :---:|
|[](https://github.com/Embodied-LLMs-Safety/Embodied-LLMs-Safety.github.io)[BadRobot：在物理世界中越狱具身 LLM](https://arxiv.org/abs/2407.20242)
Zhang, Hangtao and Zhu, Chenyu and Wang, Xianlong and Zhou, Ziqi and Yin, Changgan and Li, Minghui and Xue, Lulu and Wang, Yichen and Hu, Shengshan and Liu, Aishan and Guo, Peijin and Zhang, Leo Yu |
|[Github](https://github.com/Embodied-LLMs-Safety/Embodied-LLMs-Safety.github.io) [论文](https://arxiv.org/abs/2407.20242)|[//]: #04/21 |[](https://github.com/arobey1/robopair)
[越狱 LLM 控制的机器人](https://arxiv.org/abs/2410.13691)
Robey, Alexander and Ravichandran, Zachary and Kumar, Vijay and Hassani, Hamed and Pappas, George J |
|[项目](https://robopair.org/) [Github](https://github.com/arobey1/robopair)
[论文](https://arxiv.org/abs/2410.13691)|[//]: #04/21 |[](https://github.com/eliotjones1/robogcg)
[针对机器人视觉-语言-动作模型的对抗攻击](https://arxiv.org/abs/2506.03350)
Jones, Eliot Krzysztof and Robey, Alexander and Zou, Andy and Ravichandran, Zachary and Pappas, George J and Hassani, Hamed and Fredrikson, Matt and Kolter, J Zico |
|[Github](https://github.com/eliotjones1/robogcg) [论文](https://arxiv.org/abs/2506.03350)|[//]: #04/21 |[具身 AI 的肖申克的救赎：理解与基准测试间接环境越狱](https://arxiv.org/abs/2511.16347)
Li, Chunyang and Kang, Zifeng and Zhang, Junwei and Ma, Zhuo and Cheng, Anda and Li, Xinghua and Ma, Jianfeng |
|[项目](https://shawshankiej.github.io/) [论文](https://arxiv.org/abs/2511.16347)|[//]: #04/21 #### 运行时监控与语义对齐基准 [🔝 返回顶部](#quick-links) | 标题与作者 | 简介 | 链接 | |:--| :----: | :---:| |[](https://github.com/asimov-benchmark/asimov-benchmark.github.io)
[生成机器人准则及语义安全基准](https://arxiv.org/abs/2503.08663)
Sermanet, Pierre and Majumdar, Anirudha and Irpan, Alex and Kalashnikov, Dmitry and Sindhwani, Vikas |
|[项目](http://asimov-benchmark.github.io/) [Github](https://github.com/asimov-benchmark/asimov-benchmark.github.io)
[论文](https://arxiv.org/abs/2503.08663)|[//]: #04/21 |[SAFE-SMART：使用 STL 指标对自主机器人进行安全分析与形式化评估](https://arxiv.org/abs/2511.17781v1)
Kristy Sakano, Jianyu An, Dinesh Manocha, Huan Xu |
|[论文](https://arxiv.org/abs/2511.17781v1)|[//]: #04/21
|[](https://github.com/vla-safe/SAFE)[SAFE：用于视觉-语言-动作模型的多任务失败检测](https://arxiv.org/abs/2506.09937)
Gu, Qiao and Ju, Yuanliang and Sun, Shengxiang and Gilitschenski, Igor and Nishimura, Haruki and Itkina, Masha and Shkurti, Florian |
|[Github](https://github.com/vla-safe/SAFE) [论文](https://arxiv.org/abs/2506.09937)|[//]: #04/21 |[评估视觉-语言-动作驱动机器人的不确定性与质量](https://arxiv.org/abs/2507.17049)
Valle, Pablo and Lu, Chengjie and Ali, Shaukat and Arrieta, Aitor |
|[论文](https://arxiv.org/abs/2507.17049)|[//]: #04/21
|[视觉-语言-动作模型的置信度校准](https://arxiv.org/abs/2507.17383) Zollo, Thomas P and Zemel, Richard |
|[论文](https://arxiv.org/abs/2507.17383)|[//]: #04/21
|[视觉-语言模型能理解建筑工人吗？一项探索性研究](https://arxiv.org/abs/2601.10835) Bui, Hieu and Chodosh, Nathaniel E and Tavakoli, Arash |
|[论文](https://arxiv.org/abs/2601.10835)|[//]: #04/21标签：AI安全, Chat Copilot, Kubernetes 安全, 具身智能, 后门攻击, 学术综述, 模型鲁棒性, 视觉-语言-动作模型, 防御加固