Overview

• Google DeepMind announced Gemini Robotics-ER 1.5 for embodied reasoning alongside Gemini Robotics 1.5 as the action model.
• The ER planner perceives the environment and can call web tools like Google Search to generate step-by-step natural-language instructions.
• The action model executes those instructions with visual feedback, enabling multi-stage tasks such as sorting laundry, packing a suitcase using current London weather, and location-specific recycling.
• DeepMind reported cross-embodiment transfer, with skills moving from the ALOHA2 platform to Franka arms and Apptronik’s Apollo humanoid without specialized tuning.
• Developers can access ER 1.5 via the Gemini API in Google AI Studio, while the control model remains limited to trusted partners and testers.