Google DeepMind just dropped the next evolution of its robotics-focused AI, and it’s the kind of development that makes you rethink what machines will be capable of in the very near future. Gemini Robotics ER 2, now available in public preview through the Gemini API and Google AI Studio, is designed to serve as a high-level reasoning “brain” for robotic systems, processing video and spatial data to help robots actually understand the physical world around them.
What Gemini Robotics ER 2 actually does
The model represents a meaningful leap in what the industry calls “embodied reasoning.” It watches video feeds, understands three-dimensional space, detects objects, predicts where things are moving, and figures out how to use tools to accomplish complex tasks.
ER 2 supports multi-robot collaboration, meaning it can coordinate several machines working together on a shared objective. It also handles human-robot collaboration, so people and machines can operate as a team.
The model integrates with Gemini Robotics VLA variants, which handle direct physical control. ER 2 does the thinking, VLA handles the doing.
Google DeepMind has been building toward this moment methodically. The first Gemini Robotics model was announced in March 2025. A subsequent version, Gemini Robotics ER 1.6, launched in April 2026 and introduced instrument reading capabilities through a collaboration with Boston Dynamics. ER 2 extends those foundations with enhanced spatial and temporal reasoning, including object detection, trajectory prediction, and multi-view correspondence.
Boston Dynamics and Agility Robotics are among the companies testing and developing alongside DeepMind.
Why crypto should care about a robot’s brain
The most obvious connection point is decentralized physical infrastructure networks, commonly known as DePIN. These projects aim to build real-world infrastructure, everything from wireless networks to sensor grids to compute clusters, using token incentives instead of traditional corporate capital expenditure.
Then there’s the AI agent angle. Crypto projects building autonomous agent platforms have been among the most hyped narratives of the past year. Most of those agents exist purely in software, trading tokens or managing on-chain positions. The logical extension of agentic AI is embodied agents, software intelligence that controls physical hardware. Gemini Robotics ER 2 is essentially the blueprint for how that transition happens at the model layer.
What investors should watch
DePIN tokens tied to robotics infrastructure, sensor networks, and physical-world compute could see renewed narrative interest as mainstream AI companies validate the importance of physical AI deployment.
The risk is that Google and its hardware partners simply build the entire stack themselves, from reasoning model to direct control to cloud infrastructure, leaving no oxygen for decentralized alternatives. DeepMind’s partnerships with Boston Dynamics and Agility Robotics suggest a vertically integrated approach that could be difficult for open or decentralized ecosystems to compete with on pure capability.