one year on
Google DeepMind unveils Gemini Robotics 1.5, a two-model system for embodied reasoning and robot control
The new models enable robots to plan multi-step tasks by searching the web and transfer learned skills across different robot bodies.
Originally published Published on this site
Google DeepMind today released two new AI models designed to give robots advanced planning and reasoning capabilities. The system, called Gemini Robotics 1.5, consists of a high-level embodied reasoning model, Gemini Robotics-ER 1.5, and a vision-language-action model, Gemini Robotics 1.5. The embodied reasoning model can analyze a robot’s environment, call tools like Google Search, and generate multi-step natural language plans. The action model then translates those plans into motor commands.
The models work together in an agentic framework. In one demo, a robot searches local recycling rules online before sorting objects into compost, recycling, and trash bins. In another, it packs a suitcase based on the weather in London. The action model also generates internal reasoning steps before moving, making its decision process more transparent.
Gemini Robotics 1.5 shows cross-embodiment transfer: tasks learned on the ALOHA 2 robot work on the bi-arm Franka and Apptronik’s Apollo humanoid without retraining. Gemini Robotics-ER 1.5 is available today to developers via the Gemini API in Google AI Studio. The action model is restricted to select partners. The release includes an upgraded ASIMOV benchmark for safety evaluation.
The record
said during a press briefing that the models allow robots to 'think multiple steps ahead' before acting
said the single model can control very different robots including a humanoid, and skills learned on one robot can transfer to another
One year later — open only if you can handle spoilers
The embodied-reasoning planner opened to all developers via the Gemini API — the first Gemini Robotics model released broadly — while the full action model stayed with select partners. DeepMind kept iterating over the next year, shipping an on-device version and, by April 2026, an upgraded Gemini Robotics-ER 1.6; the planner-plus-action split and cross-embodiment transfer it introduced became common reference points for agentic robotics.
Filed
Compiled and edited by Adam Allcock.
The Weekly Replay · free by email
This week, one year ago — every Sunday.
One email each Sunday: the week's replayed AI news, with the one-year-later annotations included. Written like it's breaking — dated like it isn't.
Free · double opt-in · unsubscribe anytime · privacy