Google is turning up the ambition on robot AI again. The pitch is bigger than a flashy demo: give machines enough reasoning and dexterity to move through messy human spaces and handle jobs that used to require careful programming.

DeepMind says Gemini Robotics 2 is the “intelligence layer” for more adaptable robots, pushing beyond narrow, repetitive automation toward systems that can control “full humanoids, from feet to fingertips,” plan multi-step actions, and even work in teams.¹ In Google’s telling, this is a major leap from robots that are either pre-programmed or remotely operated, and therefore brittle when the environment changes.¹

The outside view is more measured. Axios frames the launch as part of a broader industry race to make humanoid and other robots more versatile, with models handling both fine motor control and higher-level planning.² The Verge’s emphasis is even more pointed: the headline claim is that DeepMind now has a model that can control a robot’s entire body, underscoring how central full-body coordination has become in the battle to make robots genuinely useful.³

That leaves the real tension unchanged. Google is selling breadth: robots that can “reason through every movement,” clean a cluttered room, adapt to new bodies quickly, and collaborate with other machines.¹ Skeptics will note that robot launches often sound revolutionary long before they prove durable, cheap, or safe enough for everyday deployment.

Google’s own boosters are leaning hard into the milestone. Demis Hassabis called it a system that lets robots handle tasks “that weren’t possible before, like tying delicate knots” and “team up to solve complex workflows.”Demis Hassabis (demishassabis)⁴ Sundar Pichai amplified the same message on X, a sign that Google sees robotics not as a side project but as a front-line AI battleground.Sundar Pichai (sundarpichai)⁵

What comes next matters more than the demo reel: whether Gemini Robotics 2 can move from polished lab performance to dependable work in the wild.