Google puts Gemini Robotics ER 2 into public preview

Google has made Gemini Robotics ER 2 available in public preview on its Gemini platform, bringing its embodied reasoning model for robots into enterprise production settings.

The model is designed to act as a high-level control layer for robots. It handles reasoning over video, plans tasks over longer periods and directs lower-level control systems. It can work across different robot types, including stationary arms, wheeled machines, quadrupeds and humanoids.

The release is part of an effort to place robotics tools alongside the broader cloud infrastructure businesses use to simulate, manage and deploy AI systems. Developers can connect the model to tools such as Vision-Language-Action models, navigation interfaces and custom functions, while also streaming video, audio and text into the system.

Enterprise use

One focus is physical task verification. The model can assess whether a step in a workflow has succeeded before a robot moves on, allowing it to adjust or repeat actions without restarting the full process.

It can classify task progress in binary, categorical or continuous forms. Google also said it can identify the precise video frame where a key event takes place, citing 91.3% accuracy and a mean absolute frame distance of 0.96 seconds.

Google also said the model improves spatial reasoning tasks, including visual question answering and instrument reading. According to the company, it can interpret values from more than 10 categories of instruments, including circular dials, sight glasses, digital displays, linear scales, rulers and liquid thermometers.

Another part of the offering is the use of one reasoning layer across different machine types. That could allow multiple robots to share an understanding of a workspace, divide labour and hand tasks from one machine to another.

The model also supports natural language interaction, enabling operators to issue commands in ordinary language while the robot coordinates movement and other functions through connected tools and interfaces. That could reduce reliance on technical command structures in some industrial settings.

Cloud integration

Businesses can use the Gemini platform to test robot behaviour in simulation environments before deploying systems on physical hardware. Google named MuJoCo-Warp and NVIDIA Isaac Sim among the supported environments.

Teams can also use the model's pointing, moment-finding and progress-classification functions to label demonstration videos stored in Cloud Storage and BigQuery. Those datasets can then be used to train custom Vision-Language-Action models through the Gemini platform, GKE or Cluster Director.

For live operations, workflows can be deployed with identity and security controls on the Gemini platform, while latency-sensitive local workloads such as SLAM and path planning can run on Google Distributed Cloud. Fleet telemetry can also be streamed into BigQuery.

Customer response

Several partners and customers outlined how they are using or assessing the model in robotics work.

"We've seen exponential growth in customers using AIVI-Learning every month, and Gemini Robotics ER 2 brings them even more value. Video changes what an inspection can be. Spot isn't just checking whether something looks right in a moment in time. The ability to apply AI models to video opens a whole new world of possibilities," said Marco da Silva, VP and General Manager of Spot at Boston Dynamics.

"DeepReach AI is the real-world skills engine for physical AI. Our hardware and AI platform capture how skilled people work all over the world and turn it into a data flywheel for model training. Gemini Robotics ER 2 gives us accurate task understanding of every recording, helping us find the data that matters most faster and at far greater scale," said Chris Liu, Co-founder and CTO of DeepReach.

Arm also pointed to broader implications for robotics hardware and software design.

"Advanced reasoning models have the potential to help robots interpret complex environments, plan actions and adapt to changing conditions, opening new possibilities for physical AI. Google's Gemini Robotics ER 2 model represents an important step in this direction, and we're excited to be working with Google to explore how these emerging capabilities can shape the next generation of robotics compute platforms," said Barrie Mullins, Senior Director of Arm Total Design for Physical AI at Arm.

In research settings, the model is being considered for laboratory automation that extends beyond fixed scripts and single-purpose tasks.

"We aim to accelerate materials R&D workflows so our scientists can test more ideas faster. Most lab automation today relies on rigid scripts for single, isolated tasks. By using Gemini Robotics ER 2, we are giving our robots the visual and physical flexibility to work with our existing infrastructure, whether that's reading a touchscreen on a legacy microwave reactor that has no API, balancing and loading a centrifuge, or moving samples safely through a busy lab. This is the next level we aspire to reach. It brings us much closer to self-driving labs that can run full experimental campaigns from synthesis to separation, freeing our researchers to focus on thinking up new ideas for discovering new materials," said Prof. Loh Xian Jun, Executive Director of ASTAR Institute of Materials Research and Engineering and Executive Director of MARVEL-AI.