The latest in AI, every dayAI News
← AI News

November 10, 2026 · Google DeepMind

Google DeepMind Unveils Gemini Robotics 2.5; in a Mercedes-Benz Trial, the Apollo 2 Robot Hit a 96% Success Rate

My take: A humanoid robot sorting and packing automotive parts with 96% accuracy on a real Mercedes-Benz production line isn't a lab demo anymore, it's proof that AI powered robotics is reaching real physical jobs. Google DeepMind introduced Gemini Robotics 2.5, the model giving these robots more dexterity: from using everyday tools to handling delicate objects, with tasks up to 30% more successful and 30% faster than before.

This isn't just news for robotics engineers. If your business depends on repetitive physical work (packing, sorting, assembling), this is the direction automation is heading: robots that learn tasks faster and adapt better to their environment, instead of being programmed for one exact movement. The AI here also teaches the robot to be more careful around people and obstacles, which I think is the most important detail of all: safety didn't get left behind while capability moved forward.

To me, this confirms once again that AI is a strategic tool, not magic. A robot doesn't "think" on its own, it needs a well trained and well tested model behind it, the same way any employee needs training before doing their job well.

If your company has repetitive physical processes, have you thought about which tasks could benefit from this kind of automation in the coming years, and which ones will definitely keep needing human hands?

Read at the source: Google DeepMind ↗

Want to use these tools? See the unbiased reviews or back to the news.