arXiv：cs.LG· Tianruo Rose Xu, Jiawei Ren, Yichi Yang, Zhaoxu Zheng, Lianhui Qin·· 4 小时前AI 评分54
RT-Safe：面向实时具身环境的智能体安全基准
RT-Safe: Benchmarking Agent Safety in Real-Time Embodied Environment
AI 导读
研究者提出 RT-Safe，一个在推理和执行期间世界持续演化的仿真城市基准，用于评测具身智能体在实时约束下的安全性。对八个 VLM 的测试显示，最严苛设置下仅 0.7% 的回合无安全事件完成；静态与实时评测的任务完成率分别为 91.3% 和 94.1%，但实时执行使碰撞增加 12.3 倍。论文还展示 RT-Safe 可支持离线 RL 训练，在保持较高任务完成率的同时大幅降低碰撞率。
正文
Abstract:Rapid progress in AI agents has brought growing attention to agent safety, with extensive evaluation focused on digital environments. As agents move into the physical world, embodied safety becomes increasingly important: failures can cause human injury and costly hardware damage. Beyond selecting safe actions, embodied agents must also operate under real-time constraints: the physical world does not pause while an agent reasons. As pedestrians move and vehicles approach during inference, an action that appears safe at observation time may become unsafe before execution. Real-time embodied safety therefore depends on both decision quality and decision latency. We introduce RT-SAFE, a simulated urban benchmark for evaluating embodied-agent safety under real-time constraints. RT-SAFE combines navigation tasks with moving actors, environmental hazards, and traffic rules, while allowing the world to evolve throughout inference and action execution. Across eight VLMs, agents achieve high task completion yet almost never complete safely: in the hardest setting, only 0.7% of episodes finish without a safety event. More strikingly, matched static and real-time evaluations yield task completion rates of 91.3% and 94.1%, respectively, while real-time execution increases collisions by $12.3\times$. These results reveal that standard task success can mask substantial safety failures, and that decision latency itself can become a source of physical risk. Finally, we show that RT-SAFE can support offline RL training and substantially reduce collision rates while achieving strong task completion.
| Subjects: | Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.09294 [cs.AI] |
| (or arXiv:2610.09294v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.09294 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Tianruo Rose Xu [view email]
[v1]
Wed, 7 Oct 2026 01:57:17 UTC (10,661 KB)
来源：arXiv：cs.LG · arxiv.org