Gemini Robotics 1.5 lets robots reason before they act

Google DeepMind released Gemini Robotics 1.5, a vision-language-action model, and Gemini Robotics-ER 1.5, an embodied reasoning model that plans tasks and can call tools such as Google Search. The action model can reason in natural language before moving and breaks long jobs into shorter steps. DeepMind said skills learned on one robot, such as ALOHA 2, carried over to Apptronik's Apollo humanoid and a Franka arm without retraining. The ER model was opened to developers through the Gemini API, and an upgraded ASIMOV safety benchmark accompanied the launch.

Source: Google DeepMind