Models

Google DeepMind 为 Gemini Robotics 2.5 加入机器人安全机制

Why we picked it

机器人真要走进家庭和工厂,光会干活不够,还得躲开人和危险物。DeepMind 这次把危险识别、避人、防越狱测试直接做进 Gemini Robotics 2.5 模型里。做机器人部署或安全评估的团队可以看看它的具体做法。

Google DeepMind 宣布为 Gemini Robotics 2.5 加入新的机器人安全机制,针对机器人进入家庭、工厂等真实物理场景后的风险问题。新增能力包括三个方面:让机器人自主识别环境中的危险物体、主动避让附近的人类,以及针对越狱式攻击(jailbreak)的对抗测试,防止恶意指令让机器人做出危险动作。

Image · Google DeepMind
Body

这套安全方案直接嵌在 Gemini Robotics 2.5 模型层内,而不是外挂在硬件上,意味着同一套防护可随模型部署到不同机器人本体。DeepMind 强调这是让机器人从实验室走向公共空间的前提条件。

在机器人基础模型领域,Physical Intelligence、Figure、特斯拉 Optimus 等团队都在推进真实场景部署,DeepMind 这次把安全机制作为卖点单独发布,显示出具身智能的竞争焦点正从能力演示转向可信部署,对准备引入机器人的企业和安全团队有直接参考价值。

Original · Google DeepMind

As robots move into more real-world spaces, they need greater safety measures. We’ve built in new ways for them to spot hazards, steer away from humans, and test against jailbreaks. FInd out more about Gemini Robotics 2.5 → goo.gle/4zBBUA6

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Gemini Robotics, Google DeepMind, AI安全, 具身智能.

Source attribution: curated from Google DeepMind (@GoogleDeepMind).