Google DeepMind's new AI model can control a robot's entire body
Google/DeepMind
DeepMind
Google DeepMind has released Gemini Robotics 2, which can control humanoid robots from feet to fingertips, enabling whole-body motions. The model improves dexterity with five-fingered hands and supports tasks like walking, crouching, and manipulating objects. Updated embodied reasoning (ER 2) and on-device models enhance safety, multi-robot collaboration, and offline adaptability.
Google DeepMind announced Gemini Robotics 2, an AI model that can control entire humanoid robots, including full body motions from feet to fingertips. This allows robots to walk, crouch, stretch, and manipulate objects, as shown with Apptronik's Apollo 2 robot. The model supports complex five-fingered hands for tasks like sealing a Ziploc or unscrewing a lightbulb. Updates to Gemini Robotics ER 2 improve task understanding over extended periods and enable multiple robots of different types to collaborate. The on-device model can run locally and adapt to new embodiments faster. Safety enhancements allow robots to detect humans and stop when approached too closely.
- Сокращения
- ER = Embodied Reasoning — воплощённое рассуждение
Source: The Verge —
original
