Google DeepMind Architecture Report

Gemini Robotics-ER 1.6

The most advanced embodied reasoning model, bringing agentic capabilities to robotics. Featuring unprecedented spatial reasoning, enhanced autonomy, and the safest compliance to date.

VLM
Vision-Language Model
+10%
Video Risk Detection
1.6
Model Generation

Enhanced Autonomy

Gemini Robotics-ER 1.6 introduces enhanced autonomy for robots, allowing them to reason, adapt, and respond to changes in open-ended environments. It translates natural language interactions into complex task assignments. By interpreting complex visual data and performing spatial reasoning, robots can plan actions seamlessly.

  • check_circleDeconstructs natural language commands into subtasks.
  • check_circleUnderstands object relationships and interprets dynamic scenes.
  • check_circleIntegrates with existing robot controllers to complete long-horizon tasks.

The Safest Robotics Model to Date

Safety is paramount in embodied reasoning. Gemini Robotics-ER 1.6 demonstrates superior compliance with safety policies on adversarial spatial reasoning tasks. It substantially improves the capacity to adhere to physical safety constraints compared to Gemini 3.0 Flash, accurately perceiving injury risks in text and video scenarios.