Google DeepMind has released Gemini Robotics ER 2, an AI robotics model that enables physically different robots to collaborate in real time through shared semantic understanding. The core innovation lies in temporal intelligence: the robot analyzes continuous video feeds, assesses task progress in multi-step operations, and plans actions accordingly. The model achieves 91.3% accuracy for specific event detection with a mean delay of just 0.96 seconds. Integration with the Gemini Live API enables real-time adjustments without latency, while safety improvements reduce risks for human workers. Google aims to become the operating system layer for robotics by supporting third-party hardware like Boston Dynamics and Apptronik.
Source: Read the original article
