Google DeepMind unveils Gemini Robotics 2 with whole-body humanoid control

Gemini Robotics 2 is Google DeepMind's most ambitious robotics model to date, extending control from the upper-body manipulation of prior versions to full whole-body coordination — walking, crouching, and object manipulation across a robot's entire frame. DeepMind CEO Demis Hassabis said robots can now 'reason through every movement to manage tasks that weren't possible before, like tying delicate knots — and even team up to solve complex workflows.' In a widely circulated Apptronik demo, the Apollo 2 humanoid walked across a room to place a watering can into a bin, and a separate FR3 Duo demonstration showed 20 minutes of uninterrupted, real-time tool use.

Technically, the suite spans robot form factors from tabletop arms to full humanoids, with the model reasoning about balance, coordination, and safety before executing multi-step tasks. DeepMind bundled new safety measures aimed at reducing unpredictable physical behavior — an important addition given the risk profile of untethered, whole-body machines operating near people.

The launch intensifies the 'physical AI' race, where NVIDIA anchors the infrastructure stack (Isaac simulation, robot-learning, foundation models) and startups like Apptronik and Figure supply hardware. Gemini's advantage is tight integration of a frontier multimodal reasoning model directly into motor control, letting the same system that plans a task also drive the limbs.

Skeptics on social channels pushed back on the 'physical AGI' framing, noting that curated demos rarely reflect the messiness of real homes and factories, and that whole-body autonomy raises real-world safety stakes. Still, the arrival of a whole-body model from a top lab — announced by Hassabis and amplified by Sundar Pichai — marks a clear inflection point, and readers should watch for third-party evaluations of reliability outside scripted demos.