Google launches Gemini Robotics 2 AI model that taught robots to move almost like humans
Google/DeepMind
Google DeepMind has introduced the Gemini Robotics 2 AI model, which controls all movements of a humanoid robot from feet to fingertips. The system allows robots to perform a wider range of tasks requiring whole-body coordination. The updated Gemini Robotics ER 2 model improves environmental analysis, task understanding, and safety, including better human detection.
Google DeepMind unveiled the Gemini Robotics 2 AI model, which enables full-body control of humanoid robots, coordinating movements from feet to fingertips. In a demo video, the Apollo 2 robot by Apptronik walks, squats, bends, picks up a watering can, and retrieves items from shelves. The company noted that while speed still needs improvement, this update is a crucial step toward skills for complex real-world tasks. Additionally, the Gemini Robotics ER 2 model was updated to better analyze the environment, understand instructions, perform multi-step actions, and maintain task context longer. It allows multiple robots of different types to collaborate, as shown in a garage cleanup with Apollo 2 and a dual-arm Google robot. The ER 2 model also includes enhanced safety features, detecting nearby humans and triggering a protective stop. The local Gemini Robotics On-Device Model, which runs on the robot without internet, has also been improved.
- Сокращения
- ER = Embodied Reasoning — воплощённое рассуждение
Source: 3DNews —
original
