Google DeepMind Unveils Gemini Robotics 2: Advanced AI for Dexterous, Collaborative Robots
Google DeepMind has introduced Gemini Robotics 2, its next-generation AI platform designed to enable humanoid robots to adapt to real-world environments and perform complex tasks. The system's primary advancement is full-body control, allowing robots to walk, crouch, stretch, and manipulate objects while simultaneously planning intricate actions and collaborating with other robots. Gemini Robotics 2 is comprised of three distinct AI models: Gemini Robotics 2, which translates instructions into physical movements; Gemini Robotics ER 2, responsible for reasoning, planning, and human interaction; and Gemini Robotics On-Device 2, optimized for direct execution on the robot without internet dependency. Notably, the On-Device 2 model can adapt to a new robot in just a few hours using fewer than 200 training examples. Its demonstrated capabilities include grasping small items, closing bags, tying knots, and unscrewing light bulbs. Precision tasks, such as using robotic grippers for object insertion, achieve a 90% success rate, while unscrewing a light bulb with a five-fingered robotic hand is successful 92% of the time. The robot hands in the demonstration feature 22 degrees of freedom. The reasoning model can make hundreds of decisions for tasks lasting several minutes. The platform also supports inter-robot collaboration, with different robot types effectively coordinating tasks. Google has also enhanced safety features with a new testing ground called ASIMOV-Agentic, which assesses robots' ability to refuse dangerous commands, detect uncertain situations, and halt operations when a person approaches. While inspired by Asimov's Three Laws of Robotics, the system represents a significant step toward integrating AI with physical robots for general-purpose applications, potentially serving as assistants or factory workers.
The development of Gemini Robotics 2 signifies a progression in physical AI embodiment, moving beyond simulated environments to address the complexities of real-world interaction. The emphasis on full-body control, intricate manipulation, and inter-robot collaboration highlights a strategic focus on versatile utility, from domestic assistance to industrial labor. The inclusion of on-device processing and rapid adaptation capabilities addresses critical challenges in scalability and responsiveness for robotic systems. While safety protocols are being integrated, the long-term societal implications of deploying highly capable, general-purpose robots warrant careful consideration of ethical frameworks, economic impacts on employment, and the evolving human-robot relationship in the coming decade.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.