Gemini Robotics 2 is presented as a generalist robotics model that translates natural-language goals and visual understanding into coordinated physical actionAgent orchestration coordinates AI agents, tools, people, tasks, state, and control flow so a larger workflow reaches a verified outcome.. The demonstrations emphasize whole-body balance, dexterous manipulation and adaptation rather than a robot repeating one narrowly programmed movement.
The release focuses on three capabilities. Whole-body control coordinates movement from feet to fingertips, dexterous hands handle tasks such as closing a bag or unscrewing a bulb, and separate robots use their own model copies to reason about when and how to help each other.
A high-level embodied reasoning model interprets the scene, instructions and task state, then calls a vision-language-action modelA vision-language-action model connects visual observations and language instructions to actions that a robot can perform. to produce movementA robot policy is the decision-making function that maps observations and goals to actions a robot should take in its environment.. The examples show robots dividing a garage-cleanup task, handing work from a humanoid to another robot, packing sports equipment and retrying after a failed action.
The demonstrations suggest a path toward robots that can operate in spaces designed for human bodies and respond to changing conditionsGeneralization is an AI system's ability to apply what it learned to new examples and conditions that were not part of its training experience.. They remain controlled demonstrations supplied by the developer, so they show intended capabilities rather than independent evidence of reliabilityIndependent evaluation assesses a system through reviewers whose judgment is sufficiently separate from the organization developing or promoting it. across unrestricted real-world settings.
Watch on YouTube
