Overview
- The release pairs a web‑aware planner called Gemini Robotics‑ER 1.5 with a visual‑language‑action body model called Gemini Robotics 1.5 to handle complex, multi‑step tasks.
- ER 1.5 can search the web for local rules, interpret spatial context, and produce step‑by‑step natural‑language execution plans.
- Robotics 1.5 converts those plans into movements using a think‑before‑act process that generates internal reasoning and enables explanations of its actions.
- DeepMind demonstrates cross‑morphology transfer by moving skills learned on ALOHA 2 to Apollo humanoid and Franka dual‑arm robots without retraining.
- The system incorporates layered safeguards that include pre‑action checks, adherence to existing policies, and triggers for low‑level safety subsystems.