Google DeepMind lifts the lid on a sharper robotics brain

The model specializes in reasoning capabilities critical for robotics, including visual and spatial understanding, task planning and success detection.

Gemini Robotics-ER 1.6 correctly identifies the number of hammers (2), scissors (1), paintbrushes (1), pliers (6), and a collection of garden tools which can be interpreted as a single group or multiple points. It does not point to requested items that are not present in the image — a wheelbarrow and Ryobi drill. In comparison Gemini Robotics-ER 1.5 fails to identify the correct number of hammers or paint brushes, misses the scissors altogether, hallucinates a wheelbarrow and lacks precision on plier pointing. Gemini 3.0 Flash is close to Gemini Robotics-ER 1.6, but does not handle the pliers as well. Image source: Google DeepMind.

Google DeepMind has unveiled Gemini Robotics-ER 1.6, a reasoning model meant to help robots understand the physical world with greater precision, from spatial judgment to task planning and success detection. The upgrade also improves multi-view understanding and adds instrument reading, a capability developed with Boston Dynamics, according to a blog post shared by DeepMind’s Laura Graesser and Peng Xu.

The model is meant to serve as a high-level brain for robots, able to call tools such as Google Search, vision-language-action models and third-party functions as part of its planning. DeepMind says it performs better than both Gemini Robotics-ER 1.5 and Gemini 3.0 Flash on tasks such as pointing, counting and telling whether a job has been completed.

The announcement matters beyond one lab’s technical progress. Robotics has long been constrained less by hardware than by the difficulty of making machines reason reliably in messy, changing environments; advances in embodied reasoning could therefore accelerate use cases in factories, warehouses, inspection and service robotics worldwide.

Benchmark results comparing Gemini Robotics-ER 1.6 with Gemini Robotics-ER 1.5 and Gemini 3.0 Flash models. The instrument reading evaluations were run with agentic vision enabled (except for Gemini Robotics-ER 1.5 which doesn’t support it). All other evals were run with agentic vision disabled. The single view and multiview success detection evaluations contain different examples so are not comparable. Infographic courtesy Google DeepMind.

“We are also unlocking a new capability: instrument reading, enabling robots to read complex gauges and sight glasses — a use case we discovered through close collaboration with our partner, Boston Dynamics,” Graesser and Xu said in the post.

DeepMind frames success detection as central to autonomy, arguing that robots must know not just how to act, but whether an action has worked. It also says Gemini Robotics-ER 1.6 improves multi-view reasoning, allowing the system to relate multiple camera feeds even when scenes are dynamic or partly obscured.

The company is also pitching the model as its “safest robotics model to date,” claiming superior compliance with safety policies on adversarial spatial reasoning tasks. That line will matter to industrial users, who want robots that are not merely clever but predictable. DeepMind says it is inviting developers to submit labelled images showing failure modes, suggesting the work remains a collaboration rather than a finished product.