Robotics' End Game: Nvidia's Jim Fan

Robotics is entering its "endgame" by adopting the scaling strategies that propelled large language models to success. This "Great Parallel" shifts the focus from language-heavy models to World Action Models (WAMs), which prioritize physics and motor actions as first-class citizens. By moving away from labor-intensive teleoperation toward scalable data collection methods—specifically egocentric human video and sensorized wearables—the field can overcome the physical limits of data acquisition. Integrating these with neural simulators like DreamDojo allows for massively parallel reinforcement learning, effectively turning compute into environment and data. This trajectory aims to achieve the physical Turing test, enabling robots to perform complex, dexterous tasks autonomously. With exponential technological advancement, the industry is on track to realize fully automated, "lights-out" factories and advanced scientific research platforms by 2040.