Google DeepMind's Gemini Robotics 2.5 Hits 96% Success on Mercedes-Benz Lines

Google DeepMind introduced Gemini Robotics 2.5, which it calls its most capable vision language action model so far. The model is rolling out to DeepMind's robotics partners.

DeepMind tested the model on 200 distinct tasks, such as serving food with tongs, tying a bow on a gift box and pipetting a liquid. Compared with previous baselines, it reached a 30% to 100% higher success rate depending on the task type and finished tasks 20% to 30% faster.

In a proof of concept with Apptronik and Mercedes-Benz, Apollo 2 humanoid robots running the model picked and packed automotive parts on active production lines in Kecskemét, Hungary, and Sindelfingen, Germany, with a 96% success rate. DeepMind director Kanishka Rao told Ars Technica that closing the last 4% gap will be a "steep road."

DeepMind said new safety guidance raised safe task completion by 33% around obstacles and by 43% around nearby people. It also released ASIMOV-Jailbreak, an open source benchmark of prompts that try to trick a robot into unsafe actions.