Google DeepMind 推出了 Gemini Robotics 2，這是一個全新的機器人人工智慧模型，能讓具身機器人協調全身動作。這個系統不僅讓機器人能夠行走、蹲下和操作物體，還能進行任務推理，代表了將 Gemini AI 技術延伸至實體世界的重大進展。

雖然這項發布讓 Google 延續了過去十年的機器人發展雄心，但目前機器人真正在靈巧度上仍面臨挑戰。機器人的動作依然緩慢，因為它們需要暫停思考人類可以直覺做出的決定，且從錯誤中學習的效率仍遠低於人類。

除了核心系統，Google 還發表了 Gemini Robotics ER 2，這是一個更注重安全與多步驟推理的模型。這些系統將提供給開發者及企業合作夥伴進行測試，期望未來能在多機器人協作及複雜人類環境中發揮更大的作用。

Google DeepMind unveiled Gemini Robotics 2, a new artificial intelligence model for robots that allows humanoid machines to coordinate movements across their entire bodies. The system enables robots to walk, crouch, manipulate objects, and reason through tasks, marking significant progress in extending Gemini AI technology to the physical world.

Although this release revives Google's decade-long robotics ambitions, true dexterity remains a challenge. Robot movements are still slow and deliberate because they must pause to process decisions humans make intuitively, and their ability to learn from mistakes is far less efficient than that of humans.

Alongside the core system, Google also released Gemini Robotics ER 2, a model focused on safety and multi-step reasoning. These systems will be available to developers and enterprise partners for testing, with the goal of functioning effectively in multi-robot collaborations and complex human environments.