Google DeepMind launches Gemini Robotics 2.5 VLA model

Google DeepMind introduced Gemini Robotics 2.5, its most capable vision-language-action model, improving robot dexterity, reliability, and learning speed across broader real-world tasks. In a production-line proof of concept with Apptronik’s Apollo 2 humanoid at Mercedes-Benz facilities, it achieved a 96% success rate in parts kitting, indicating progress toward practical industrial deployment of generalist robot models.