Google DeepMind's vision-language-action robot model
Grounded in available product and source data
Gemini Robotics 2.5 is a vision-language-action model from Google DeepMind for robots. It widens the set of skills a robot can learn and speeds up how quickly it learns them, which the team says improves dexterity, reliability and safety in real settings.
In a proof of concept on Mercedes-Benz production lines, the models helped Apptronik's Apollo 2 humanoid pick and pack automotive parts with a 96% success rate. The work is rolling out to robotics partners, alongside a new benchmark for adversarial jailbreaks.