Google DeepMind has released Gemini Robotics-ER 1.6, a specialized AI model designed to help robots understand and act in physical environments by improving their ability to interpret visual information, plan complex tasks, and identify safety hazards. The new model represents a significant step toward embodied AI, where machines can bridge the gap between digital intelligence and real-world action .

What Makes This Robotics Model Different From General-Purpose AI?

Unlike large language models (LLMs) designed primarily for text-based conversations, Gemini Robotics-ER 1.6 is purpose-built for robots that need to understand and manipulate physical objects. The model functions as a reasoning layer that processes visual inputs from cameras and translates them into actionable decisions. This is particularly important in industrial settings where robots must work autonomously without constant human supervision .

The model can interact with external tools, including search functions and vision-language-action systems, to support task execution. This means a robot doesn't just see an object; it can reason about what to do with it, whether the action is safe, and when the task is complete .

How to Deploy Gemini Robotics-ER 1.6 in Your Operations

• Access via Gemini API: Developers can integrate the model directly into robotics systems through Google's Gemini API, with example workflows and developer tools provided to streamline implementation.
• Use Google AI Studio: The model is also available through Google AI Studio, offering a no-code interface for testing and prototyping robotics applications before full deployment.
• Leverage Multi-View Perception: Combine inputs from multiple cameras, such as overhead and wrist-mounted views, to build a complete understanding of dynamic or partially obscured environments in your facility.

What New Capabilities Does This Model Add?

Google DeepMind highlighted several key improvements in Gemini Robotics-ER 1.6 that directly address real-world robotics challenges :

• Spatial Reasoning and Object Understanding: The model improved its ability to identify, count, and locate objects with fewer errors, such as falsely identifying objects that are not present in a scene.
• Pointing and Relational Reasoning: Uses spatial "pointing" as an intermediate reasoning step to understand relationships, trajectories, and constraints in complex environments.
• Task Planning and Success Detection: Determines whether a task has been completed successfully, allowing robots to decide whether to retry an action or move to the next step in a workflow.
• Multi-View Perception: Combines inputs from multiple cameras to build a more complete understanding of dynamic or partially obscured environments.
• Instrument Reading: Interprets gauges, thermometers, and sight glasses, a capability developed in collaboration with Boston Dynamics for inspection and monitoring tasks in industrial settings.

The instrument-reading capability is particularly noteworthy because it addresses a practical challenge in industrial automation. Robots like Boston Dynamics' Spot can now capture images of equipment and interpret readings autonomously using what Google calls "agentic vision," a combination of visual reasoning and intermediate computational steps such as zooming into images and estimating measurements .

How Does Safety Factor Into This New Model?

Safety improvements are a central focus of Gemini Robotics-ER 1.6. The model shows improved adherence to physical constraints and better identification of potential hazards. Google DeepMind evaluated the system on tasks involving safety instruction following and risk detection in both text and video scenarios. The results were significant: on these safety-focused tasks, Gemini Robotics-ER models improved over baseline Gemini 3.0 Flash performance by 6 percent in text-based scenarios and 10 percent in video-based scenarios when perceiving injury risks accurately .

"Capabilities like instrument reading and more reliable task reasoning will enable Spot to see, understand and react to real-world challenges completely autonomously," said Marco da Silva, Vice President and General Manager of Spot at Boston Dynamics.

Marco da Silva, Vice President and General Manager of Spot at Boston Dynamics

This partnership between Google DeepMind and Boston Dynamics demonstrates how the model is being tested in real-world conditions. Spot, a quadruped robot used for inspection and monitoring tasks, can now operate with greater autonomy and safety awareness .

Why Does This Matter for the Future of Robotics?

The release of Gemini Robotics-ER 1.6 signals Google DeepMind's commitment to connecting advances in AI models with practical robotics applications. As Google noted in the announcement, "For robots to be truly helpful in our daily lives and industries, they must do more than follow instructions; they must reason about the physical world" .

As Google

This reasoning capability is what separates modern embodied AI from earlier generations of robots that relied on pre-programmed instructions. A robot equipped with Gemini Robotics-ER 1.6 can navigate a complex facility, interpret equipment readings, identify hazards, and adapt to unexpected situations, all without human intervention. The model is available now through the Gemini API and Google AI Studio, making it accessible to developers and organizations looking to deploy autonomous robotics systems .

The broader context is important: Google Gemini has grown to 750 million monthly active users across its consumer and enterprise products, with 2.4 million developers building on the Gemini API . By extending Gemini's capabilities into robotics reasoning, Google is expanding its AI platform into new domains where physical understanding and autonomous decision-making are critical.