AI robots are passing factory tests — what that means for anyone thinking about automation
Google's AI gave robots a 96% success rate sorting car parts at real Mercedes-Benz factories, showing that AI-powered robots are now crossing from demos into actual industrial work.
- 96%
- success rate at real Mercedes-Benz factories
- 200+
- distinct tasks tested
- 20–30%
- faster than the previous version
- Nov 10, 2026
- date of the DeepMind release
Google DeepMind's Gemini Robotics 2.5 just did something most robot demos skip: it ran on a real production line. At two Mercedes-Benz factories — one in Kecskemét, Hungary, and one in Sindelfingen, Germany — an Apollo 2 humanoid robot sorted metal car parts into trays, in the exact order each vehicle needs them. It got the job right 96% of the time, which DeepMind calls promising progress toward real industrial use.
That number matters because the hard part of industrial automation has never been making a robot work in a controlled lab. The challenge is making it work in a real factory — changing light, slightly different parts each day, workers moving nearby. The same AI model that learned by practicing 200 different tasks, from folding clothes to packing food to hanging mugs, transferred to a live factory with very little adjustment. That is the real milestone.
DeepMind reports that the new model completes tasks 20–30% faster than its previous version and achieves 30–100% higher success rates across different kinds of work. The model is currently rolling out through partners — Apptronik, Agile Robots, and Boston Dynamics — with no public API yet. It runs alongside a separate planning model that tracks the overall workflow while Gemini Robotics 2.5 controls the precise hand and arm movements.
For anyone running operations or building automation products, the signal is straightforward: the hardest problem in physical automation — handling real-world variation — is being solved. The same advances making AI better at writing and coding are now making robots more reliable on the floor. If you work in manufacturing, logistics, or any work that involves picking, sorting, or packing physical things, the question is no longer whether this technology works. The question is when to get started.