Google DeepMind is l aunching two new AI models designed to help robots “perform a wider range of real-world tasks than ever before.” The first, called Gemini Robotics, is a vision-language-action model capable of understanding new situations, even if it hasn’t been trained on them.