The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

Google DeepMind introduced Gemini Robotics ER 2, a model intended to function as a robot’s central decision-making unit. It supports real-time spatial reasoning, multi-step task planning, and collaboration between different robots. Developers can access the model via the Gemini API, Google AI Studio, or the Gemini Enterprise Agent Platform to build physical AI agents. The model processes continuous video feeds to track progress and adapt actions dynamically, improving reliability in real-world environments.

Gemini Robotics ER 2 enhances task execution by enabling robots to self-correct and generalize to new situations. It integrates with lower-level vision-language-action models and can call external tools such as Google Search or user-defined functions. The model supports bidirectional streaming for low-latency operations, allowing fluid task orchestration without pauses. A demonstration with Boston Dynamics’ Spot robot shows it fetching objects based on natural language commands, with code available on GitHub.

The update introduces significant improvements in video understanding and progress tracking, addressing a key challenge in robotics: determining task completion. Gemini Robotics ER 2 classifies task progress into five stages and identifies critical moments in video feeds to switch tasks precisely. It achieves 57.4% accuracy in progress classification and 91.3% in moment-finding, outperforming prior models while maintaining lower computational costs and higher execution speeds.

Gemini Robotics ER 2 also advances multi-robot collaboration, allowing diverse robots to work together through shared semantic understanding. The model demonstrates strong performance in safety benchmarks, including halting operations near humans and resuming only when safe. It outperforms previous versions and competing models in safety instruction adherence and human proximity detection, contributing to safer physical AI deployments.