Google DeepMind’s new model represents a major leap toward robots that can generalize, effortlessly adapting to new tasks, objects, and environments with minimal training. Built on Gemini’s world understanding, the model tackles a wide range of tasks, even those it has never encountered before. It outperforms other leading vision-language-action models in generalization benchmarks. The team envisions this breakthrough leading to robots that are far more useful and require less detailed training for each task. This advancement could make robots more adaptable and efficient in real-world scenarios. Watch robots pack a lunch, fold laundry, and fold origami as part of their enhanced capabilities.
Gemini Robotics: Bringing AI to the physical world
Gemini Robotics: Generality in action
Gemini Robotics uses Gemini’s world understanding to adapt to new situations, objects, and environments, solving tasks it hasn’t encountered in training. It outperforms other vision-language-action models, achieving over double the performance on generalization benchmarks.
Gemini Robotics brings
Gemini Robotics, built on Gemini 2.0, is highly interactive, understanding and responding to a wide range of natural language commands. It adapts to changes in its environment and instructions, allowing seamless collaboration with people in various settings.
Gemini Robotics: Dexterous skills
Gemini Robotics excels in dexterity, performing complex tasks like origami folding and packing snacks, which are challenging for most robots.
DeepMind also prioritizes safety, releasing new frameworks and datasets to ensure responsible AI development in robotics.
Read more here.
