DeepMind's new AI gives humanoids whole-body control

Robots are moving past tabletop tricks and learning how to use their whole bodies.

Google DeepMind announced Gemini Robotics 2, a trio of models designed to give physical hardware full-body control, finer dexterity, and teamwork skills. The update includes a motor-control model that operates humanoids from feet to fingertips, a reasoning model that plans multi-minute tasks, and an on-device model that adapts to new robot bodies in a few hours.

Why it matters: Most robots rely on pre-programmed scripts and struggle outside predictable settings. These models let humanoids crouch, walk, tie knots, and coordinate with other bots to handle real-world chores without getting stuck.

If you want to build with it, the reasoning model (Gemini Robotics ER 2) is live on Google AI Studio, while the motor-control models are currently open to early-access partners.

They are still pretty slow on their feet, but at least your future helper won't fall over picking up a watering can.