Google DeepMind unveils Gemini Robotics 2 with whole-body humanoid control and multi-robot teamwork

Google DeepMind has introduced Gemini Robotics 2, an update to its intelligence layer for robots that adds whole-body control of humanoids, finer dexterity in hands and grippers, and the ability for different robots to work together. The system is delivered as three models: Gemini Robotics 2, a vision-language-action (VLA) model that converts vision and language input into motor control; Gemini Robotics ER 2, an embodied-reasoning model that acts as the robot's high-level brain; and Gemini Robotics On-Device 2, an efficient VLA that runs locally on a robot without network access.

While previous versions controlled a humanoid's upper body for tabletop tasks, Gemini Robotics 2 can direct entire humanoid bodies — from feet to fingertips. On Apptronik's Apollo 2 humanoid, the team demonstrated the model responding to an instruction to 'put the watering can into the green bin in the bottom shelf': the robot walked to the table, picked up the can, took a few steps to the shelves, and placed it in the destination.

On dexterity, Gemini Robotics 2 can control the five-fingered, 22-degree-of-freedom SharpaWave hand on Apollo 2 to complete delicate actions such as tying knots or sealing a ziplock bag. It can also operate standard two-fingered parallel grippers on a Franka Duo platform for tasks like tight packing.

The embodied-reasoning model Gemini Robotics ER 2 acts as the high-level planner: it observes the room, reasons about the steps needed, coordinates with the VLA to carry out each action, and tracks progress. In this update, it can execute longer sequences lasting several minutes and involving hundreds of decisions, and can pinpoint when steps begin and end. It also introduces multi-robot collaboration, letting different robots communicate and work together on workflows a single robot could not complete alone.

For applications where network latency or connectivity is a problem, Gemini Robotics On-Device 2 runs locally on the robot and can adapt to a completely new robot body with a few hours of data. Gemini Robotics ER 2 is available on Google AI Studio and in private preview on the Gemini Enterprise Agent Platform; the VLA and On-Device models are available to early-access partners.