HOW WE BUILT THE NEW FAMILY OF GEMINI ROBOTICS MODELS

Summary from the TLDR Web Dev newsletter

Google DeepMind has developed a new family of Gemini Robotics models, built upon Gemini 2.0, to let robots perform complex tasks like preparing food and playing games. These multimodal models add physical action capabilities to Gemini's existing abilities by fine-tuning with robot-specific data, allowing robots to understand and react to new objects and instructions without retraining. The Gemini Robotics-ER model focuses on embodied reasoning, while the Gemini Robotics model is best for dexterity.