Google DeepMind's Gemini Robotics 2.5 Demonstrates High Success Rate on Mercedes-Benz Assembly Line
谷歌VLA机器人杀入奔驰车间！96%成功率引爆2026具身智能热搜
According to a report dated November 11, 2026, Google DeepMind's Gemini Robotics 2.5 model is being tested at the Mercedes-Benz assembly plant in Stuttgart. The system reportedly achieves a 96% success rate in tasks such as grasping brake calipers with tight tolerances during continuous operation. The article describes the model's safety features, including real-time guidance and defense against adversarial prompts, while noting that physical variability remains a challenge for full deployment.
Source: Chinese robotics — Robot learning research · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
96%
Reported trials
View original evidence
【2026年11月11日深夜，全球机器人圈集体刷屏】——不是科幻预告片，不是概念发布会，而是真·钢铁打工人上岗实录：Google DeepMind最新发布的Gemini Robotics 2.5，正站在斯图加特奔驰总装线的传送带旁，用灵巧手稳稳抓起一颗毫米级公差的制动卡钳，转身、定位、轻放托盘——动作行云流水，成功率高达96%。
Open source S2training duration
11
Reported success rate
33%
View original evidence
Gemini Robotics 2.5首次嵌入‘推理时引导机制’（Inference-time Guidance），相当于给机器人装了一套实时OS内核：当机械臂靠近传送带护栏时，系统自动微调轨迹曲率；当产线工人突然走入工作半径1.2米内，末端执行器立即降速至0.1m/s并启动声光预警——测试数据显示，障碍物场景安全成功率↑33%，人员邻近场景↑43%。
Open source S14failure cost implication
400
View original evidence
💡 三、热搜背后的真相：大模型不是万能钥匙，但它是第一把万能钥匙坯 当然，也有清醒的声音在评论区敲黑板：‘96%很震撼，但剩下4%的失败，可能对应400台车的返工成本。
Open source S19
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
所以当Apollo2机器人搭载Gemini Robotics 2.5，在连续72小时无监督运行中稳定达成96%零部件识别-抓取-定位-放置闭环成功率时，现场工程师当场掏出手机录下视频发到LinkedIn，配文只有一句：‘We’re not simulating anymore. We’re shipping.’（我们不再仿真了，我们正在交付。
Open source S10
【2026年11月11日深夜，全球机器人圈集体刷屏】——不是科幻预告片，不是概念发布会，而是真·钢铁打工人上岗实录：Google DeepMind最新发布的Gemini Robotics 2.5，正站在斯图加特奔驰总装线的传送带旁，用灵巧手稳稳抓起一颗毫米级公差的制动卡钳，转身、定位、轻放托盘——动作行云流水，成功率高达96%。
Open source S2
训练周期从3个月压缩到11天，真机磨损降低76%——这才是谷歌敢说‘几周后训完就上奔驰线’的底气。
Open source S21
Gemini Robotics 2.5首次嵌入‘推理时引导机制’（Inference-time Guidance），相当于给机器人装了一套实时OS内核：当机械臂靠近传送带护栏时，系统自动微调轨迹曲率；当产线工人突然走入工作半径1.2米内，末端执行器立即降速至0.1m/s并启动声光预警——测试数据显示，障碍物场景安全成功率↑33%，人员邻近场景↑43%。
Open source S14
结果Gemini 4 Argon以99.5%防御率碾压前代——这意味着，它不仅懂怎么干活，更懂‘什么活绝对不能干’。
Open source S17
💡 三、热搜背后的真相：大模型不是万能钥匙，但它是第一把万能钥匙坯 当然，也有清醒的声音在评论区敲黑板：‘96%很震撼，但剩下4%的失败，可能对应400台车的返工成本。
Open source S19
Source：Chinese robotics — Robot learning research · sohu.com