A groundbreaking development in robotics and AI has emerged: a robot capable of learning Lego assembly from a single demonstration. This generalist model, developed by GeneralistAI, controls two robotic hands in real-time using a neural network running at 100Hz. The system demonstrates remarkable capabilities, including one-shot assembly, dexterous manipulation, and integrated reasoning. The robot can replicate Lego structures after seeing them just once, showcasing sub-millimeter precision in grasping, nudging, and aligning Lego bricks. What sets this apart is its ability to combine high-level planning (“what to build”) with low-level execution (“how to build it”) in a single model, without task-specific code or custom rules.
