Google DeepMind shipped Gemini Robotics 2 on July 30, and the headline is simple: last year’s model only drove a robot’s upper body. This one runs the whole machine — walking, crouching, reaching, gripping — while it reasons through a multi-step task in real time.
What it actually does
It’s a vision-language-action model, not a chatbot. Point a humanoid at a job and Gemini’s reasoning plans the steps, then coordinates the entire body to execute. The dexterity demos are the part people are staring at: screwing in a light bulb, tying knots — the finicky, finesse-heavy motions robots usually fumble. It can also sync multiple robots in a shared space to finish work one machine can’t.
Why it matters
The real trick is zero-shot generalization. Give it a task it was never trained on and it adapts on the fly, reportedly porting to a new bi-arm robot in hours instead of months of robot-specific tuning. Bloomberg and The Robot Report covered it the same day; Hacker News put it on the front page at 92 points.
Access is going to robotics developers and partners through the Gemini Robotics platform — no public self-serve API yet.
You Might Also Like
- Google Deepmind Aletheia Just Solved Math Problems Nobody Could Heres why That Matters
- Google Lyria 3 Just Turned Gemini Into a Music Studio and im Weirdly Into it
- Gemini Canvas in ai Mode Google Just Turned Search Into a Creative Workspace
- Google Releases Gemini Embedding 2 one Vector Space for Text Images Video and Audio
- Google is Deploying Gemini ai Agents to 3 Million Pentagon Employees Heres What That Means
Top AI Product 
Leave a comment