Descriptions:
Google has unveiled Gemini Robotics 2, a generalist AI model designed to control full humanoid robot bodies rather than repeating narrowly programmed movements. The video covers the three pillars of this release: whole-body coordination, advanced dexterous manipulation, and multi-robot collaboration — demonstrated through robots named Apollo and Duo performing tasks like packing ziplock lunch bags, unscrewing light bulbs, tying knots in trash bags, and jointly organizing tools into a kit.
The key engineering insight highlighted is that Gemini Robotics 2 controls all 22 joints of a robotic hand simultaneously, requiring the model to translate high-level reasoning into precise physical motion in real time. Researchers explain how a high-level reasoning model breaks tasks into sub-goals and dynamically hands off control between robots, with each running its own copy of the model stack and coordinating through shared reasoning rather than a single centralized network.
The video situates this work within the broader argument that AI is the missing piece of the robotics puzzle — enabling generalist robots that can handle the messy, unstructured complexity of real-world environments. Google researchers describe ambitions around hazardous waste handling and other tasks where sending a robot instead of a human would reduce risk, signaling that Gemini Robotics 2 is positioned as a platform for real-world deployment, not just lab demonstration.
📺 Source: TheAIGRID · Published July 31, 2026
🏷️ Format: News Analysis
