[Guosen Embodied Intelligence Team] Google Releases Gemini Robotics 2, Embodied Intelligence Ushers in 'Android' Moment — 0731

English summary

Event: On July 30, Google released Gemini Robotics 2, becoming the intelligent foundation supporting the next generation of truly adaptive robots. Gemini Robotics 2 for the first time extends the model's control range from the upper body to the entire humanoid robot, achieving three core capabilities: intelligent whole-body motion control, advanced dexterous manipulation, and multi-robot collaborative operations. Core architecture: ①Gemini Robotics 2 — Execution: A cross-embodiment VLA model that converts visual and language inputs into action instructions, adaptable to various robots; ②Gemini Robotics ER 2 — Reasoning: An embodied reasoning model, essentially a VLM, enabling human-robot interaction, understanding the physical world, long-horizon task planning, and multi-robot collaboration; ③Gemini Robotics On-Device 2 — On-device: A lightweight on-device VLA model optimized for local offline robot operation, capable of fast cross-embodiment adaptation after a few hours of training.…