Google Launches Gemini Robotics ER 2: Its Most Advanced AI Model for Robotics

Google has released Gemini Robotics ER 2, an artificial intelligence model designed to enhance robotic coordination, communication, and fine-motor dexterity. According to the company, the model enables robots to perform complex physical tasks such as tying knots and coordinating movements with other units through a shared intelligence framework.

Gemini Robotics ER 2 Technical Capabilities

The ER 2 model focuses on bridging the gap between high-level reasoning and low-level physical execution. Google states that the system allows robots to process multimodal inputs to understand their environment and execute precise manual tasks. One specific capability highlighted by the company is the ability of robots to tie knots, a task that requires significant tactile feedback and spatial awareness.

Beyond individual dexterity, the model introduces a coordinated operational structure. Google describes this as a hive-mind approach, where multiple robots can converse and synchronize their actions in real time to complete collective goals. This coordination reduces the need for individual programming for every specific movement, as the model manages the distribution of tasks across the robotic group.

Integration of Large Language Models in Robotics

Gemini Robotics ER 2 leverages the reasoning capabilities of the Gemini family of models to translate natural language instructions into robotic actions. This integration allows users to communicate with robots using conversational language, which the AI then decomposes into a series of executable physical steps.

By using a foundation model for robotics, Google aims to move away from traditional “hard-coded” robotics. Instead of writing specific scripts for every possible scenario, the ER 2 model uses generalized learning to adapt to new objects and environments. This approach allows the robots to infer the properties of an object—such as whether it is fragile or flexible—and adjust their grip and force accordingly.

Industry Context and Robotic Coordination

The shift toward coordinated robotic systems addresses a long-standing challenge in automation: the “handoff” problem. In industrial and laboratory settings, robots often struggle to transfer items or collaborate on a single object without precise, pre-defined synchronization. The ER 2 model’s ability to facilitate communication between units suggests a move toward more fluid, autonomous collaboration in unstructured environments.

This development follows a broader trend of applying Large Multimodal Models (LMMs) to physical hardware. By combining vision, language, and action (VLA) into a single model, Google is attempting to create a more intuitive interface for human-robot interaction, reducing the technical barrier for deploying robots in non-factory settings.