Google DeepMind Gemini Robotics 1.5 Pushes Robots Toward Physical Agents

Google DeepMind introduced Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 as a step toward physical agents: AI systems that can understand a real environment, create a plan, use digital tools when needed and then translate that plan into robot actions.

The system separates high-level embodied reasoning from low-level action. Gemini Robotics-ER 1.5 acts like a planning layer that understands scenes, reasons about spatial relationships, estimates progress and can call tools such as search or user-defined functions. Gemini Robotics 1.5 then turns natural-language step instructions and visual context into motor commands.

The practical implication is important for long-horizon tasks. Instead of asking a robot to repeat a narrow preprogrammed behavior, a user could ask it to sort objects by local recycling rules, organize laundry by color, or perform a task that requires multiple decisions before movement begins.

DeepMind also highlighted cross-embodiment learning, showing that skills learned on one robot form can transfer to others such as ALOHA-style arms, Franka arms and Apptronik's Apollo humanoid. That matters because the robotics field has long struggled with the cost of retraining every behavior for every body.

For the industry, the announcement strengthens the idea that the next robotics race will be fought as much in foundation models and embodied reasoning as in motors, batteries and dexterous hands.

The next robotics race is becoming a race for physical intelligence.

— RobotsHome Editorial