Google DeepMind's Gemini Robotics 2 represents a significant leap in AI for robots, moving beyond pre-programmed tasks to enable intelligent whole-body control, fine dexterity, and multi-robot collaboration. This new intelligence layer allows robots to reason through every movement, facilitating complex actions like cleaning cluttered rooms and teaming up to increase efficiency.

Key advancements include Gemini Robotics 2 (VLA model) for full humanoid control and dexterous manipulation, Gemini Robotics ER 2 (embodied reasoning model) for multi-step task planning and human communication, and Gemini Robotics On-Device 2 for efficient local operation and rapid adaptation to new robot bodies. The system can now control entire humanoid robots, execute delicate tasks with advanced dexterity, and coordinate multiple robots for complex workflows.

Furthermore, Gemini Robotics 2 prioritizes safety with new benchmarks like ASIMOV-Agentic, enhancing the robot's ability to refuse unsafe actions and proactively seek human intervention. This development is a crucial step towards creating adaptable, general-purpose robots capable of operating safely and effectively in real-world environments.