Overview

• Google DeepMind introduced a two-model Gemini Robotics 1.5 system that separates reasoning from control to enable multi-step planning.
• Gemini Robotics-ER 1.5 can use web tools like Google Search to gather context and produce natural-language plans for real-world tasks.
• Gemini Robotics 1.5 converts those plans into actions with visual guidance and a think-before-acting process for step-by-step execution.
• The approach supports cross-embodiment learning, with skills carrying over across ALOHA2, Franka, and Apptronik’s Apollo humanoid without special retuning.
• Demos showed laundry sorting, suitcase packing using current London weather, and location-specific recycling, as safety, privacy, and reliability testing continues before wider rollout.