Google DeepMind’s new AI model can control a robot’s entire body

Google DeepMind has introduced the latest version of its Gemini Robotics AI model, which now supports whole-body motions for humanoid robots, ranging from feet to fingertips. This update enables robots like Apptronik’s Apollo 2 to walk, crouch, stretch, and handle complex manipulation tasks such as sealing bags or unscrewing lightbulbs. Additionally, the company updated its Gemini Robotics ER 2 vision-language model for better spatial reasoning, extended task duration, multi-robot collaboration, and enhanced safety features like human proximity detection. The company also improved its on-device model, allowing it to run locally and adapt quickly to robots with drastically different shapes and sensors.

(Source：The Verge)