A new robotics control model
Google DeepMind introduced Gemini Robotics 2.5, a model designed to control robots. The company tested it with robot maker Apptronik on active Mercedes-Benz production lines in Hungary and Germany.
What it did on the line
During the tests, Apptronik's humanoid robot Apollo 2 took on the task of sorting metal parts. According to DeepMind, the robot reached a 96 percent success rate.
Why it matters
Such pilots are notable for showing how reliable robotic AI models are outside the lab, on real factory lines. Repetitive tasks like sorting metal parts are among the first targets for automation.
Limits
The shared details are limited to a single task and pilot runs at two sites. It was not disclosed under what conditions the success rate was measured or when production deployment might follow.
