Google DeepMind Releases Gemini Robotics ER 2 for Advanced Embodied Reasoning
Google DeepMind has launched Gemini Robotics ER 2, its most advanced embodied reasoning model designed to act as a high-level brain for robots. The model introduces support for continuous video stream analysis, parallel reasoning to reduce latency, and multi-robot collaboration. This model significantly improves robotic efficiency by allowing robots to track task progress in real-time, self-correct errors, and coordinate with other machines. By achieving sub-second latency at a lower computational cost, it brings AI-driven robots closer to safe, real-world deployment. Gemini Robotics ER 2 achieves a 91.3% accuracy rate in moment-finding tasks and runs four times faster than larger model classes. It integrates with the Gemini Live API for low-latency streaming and is now available for developers via the Gemini API and Google AI Studio.
## BACKGROUND
Embodied reasoning is a cognitive process where an AI agent or robot grounds its high-level planning in the physical constraints of its body and environment. In robotics architectures, high-level reasoning models act as the "brain" that plans tasks, which are then translated into physical movements by lower-level Vision-Language-Action (VLA) models.