Gemini Robotics 2: Google Puts a Brain in the Whole Body
DeepMind shipped Gemini Robotics 2 on July 30, and the headline isn't better hands, it's a whole body that knows what it's doing. Version 1 could work an upper body. Version 2 does full humanoid control, feet to fingertips, and it comes as three models instead of one.
The split is the smart part. Gemini Robotics 2 is the VLA, the thing that turns what the robot sees and hears into motor commands, and it works across different hands and grippers. Gemini Robotics ER 2 is the high-level brain, the embodied-reasoning model that plans multi-step tasks, talks to humans, and coordinates several robots working together for minutes at a stretch. And Gemini Robotics On-Device 2 runs locally on the robot and can adapt to a brand-new body in a few hours instead of a training run.
Why it matters: the agent story has lived on a screen for two years, an LLM clicking buttons and calling tools. This is the same playbook pointed at atoms. Same reasoning core, same tool-use loop, except now the tool is a leg. Multi-robot collaboration and self-correction are the tells that Google thinks the physical world is the next agent arena, not a demo.
Right now ER 2 is on Google AI Studio and in private preview inside the Gemini Enterprise Agent Platform. The VLA and the on-device model are locked to early-access partners, so you can plan against the brain today and wait on the body. Details at deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots.
← Back to all articles
The split is the smart part. Gemini Robotics 2 is the VLA, the thing that turns what the robot sees and hears into motor commands, and it works across different hands and grippers. Gemini Robotics ER 2 is the high-level brain, the embodied-reasoning model that plans multi-step tasks, talks to humans, and coordinates several robots working together for minutes at a stretch. And Gemini Robotics On-Device 2 runs locally on the robot and can adapt to a brand-new body in a few hours instead of a training run.
Why it matters: the agent story has lived on a screen for two years, an LLM clicking buttons and calling tools. This is the same playbook pointed at atoms. Same reasoning core, same tool-use loop, except now the tool is a leg. Multi-robot collaboration and self-correction are the tells that Google thinks the physical world is the next agent arena, not a demo.
Right now ER 2 is on Google AI Studio and in private preview inside the Gemini Enterprise Agent Platform. The VLA and the on-device model are locked to early-access partners, so you can plan against the brain today and wait on the body. Details at deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots.
Comments