Google DeepMind releases Gemini Robotics 2.5 for robot dexterity

Google DeepMind introduced Gemini Robotics 2.5, a vision-language-action model that it says can learn a broader range of skills and acquire them faster than earlier versions. The company evaluated it on more than 200 tasks, including using tools, tying a bow on a gift box and handling flexible objects, and reports success rates 30% to 100% higher than previous baselines, with tasks finished 20% to 30% faster. It works with several robot types, including Apptronik's Apollo 2 humanoid and the Franka Duo platform.

In a proof-of-concept on Mercedes-Benz production lines in Hungary and Germany, the model guided Apollo 2 through automotive kitting with a 96% success rate, sorting parts into trays in the order required for each vehicle. DeepMind also released ASIMOV-Jailbreak, an open-source benchmark of prompts that disguise unsafe requests. Gemini Robotics 2.5 is rolling out to the company's robotics partners.