Google DeepMind has introduced Gemini Robotics 2, a new version of its robotics AI that, according to Memeburn, gives robots "full-body intelligence" — whole-body control, dexterous hands, and the ability to plan as a team.
The shift matters because most robot demos you've seen are narrower than they look. A machine that can pick up a single object on a table is solving a much smaller problem than one that has to keep its own balance while reaching, stepping, and adjusting to a shifting load. Whole-body control means the AI is coordinating the entire machine at once rather than treating the arm, the hands, and the legs as separate systems.
A report carried on MSN describes the same release as letting humanoid robots walk, handle delicate tasks, and coordinate with other robots — while, in its words, "staying safer around humans." That last point is the quiet headline. Robots that operate near people have to fail gracefully, and safety behavior is one of the harder things to get right in a system driven by a general-purpose model.
The teamwork element is the other notable claim. Both Memeburn and the MSN report describe multi-robot planning, meaning machines that divide up a job rather than each running an isolated script.
Worth noting: the available coverage summarizes capabilities rather than publishing benchmark results, so there are no independent performance numbers here yet, and no detail on which hardware platforms this runs on or when it reaches customers.
It matters because the bottleneck in robotics has long been software, not motors — and a general AI model that can run a whole body, hands included, is the clearest path yet from lab demos to machines that do useful physical work alongside people.