Analysis
Google DeepMind unveiled Gemini Robotics 2 this week, extending its robotics model to control an entire humanoid robot's body -- walking, crouching, balancing and manipulating objects -- rather than the arms-and-upper-body focus of its predecessor.
The new model treats video as a first-class input alongside text and sensor data, meaning robots running it can interpret and act on raw visual scenes directly instead of depending only on structured commands or pre-mapped environments. Google is also pitching the model for coordinating multiple robots on a shared task, aiming the initial use case at warehouse and logistics deployments rather than consumer-facing humanoids.
The launch lands in the same week the US added foreign-made humanoid and quadruped robots to the FCC's national-security Covered List, a policy shift that pushes US robotics makers toward domestic hardware. That creates an opening for whichever company supplies the default AI "brain" running on top of that hardware, and Google's simultaneous push into full-body robot control positions Gemini as a serious candidate for that role alongside builders like Agility Robotics and Figure.
Alphabet's ability to ship the same model generation into both a research robotics platform and, separately this same week, a consumer-facing Waymo product update, reflects a level of vertical integration between model research and physical deployment that few competitors can match without owning both an AI lab and a robotics or autonomy business outright.
What to watch: which hardware makers actually adopt Gemini Robotics 2 versus building proprietary control stacks, and whether Google's timing relative to the US robot-import ban helps or complicates its robotics partnerships with companies that also sell into China.
