In 1979, Hans Moravec’s Stanford Cart did something genuinely new: it “successfully crossed a chair-filled room without human intervention,” using a camera that slid along a rail to capture stereo images and a computer to plan a path around obstacles. As a demonstration of autonomous mobile robotics, it was a landmark.
As a practical machine, it was almost comically slow. The Cart did not drive smoothly. It would roll forward a short distance, stop, take a set of pictures, and then grind through the vision and planning computation before deciding where to move next. Crossing a single cluttered room could take hours. The bottleneck was not the wheels or the motors - it was perception. Seeing the world well enough to know where the chairs were turned out to be the hard part.
That experience seeded one of the most quoted ideas in robotics, later associated with Moravec himself: the tasks that feel effortless to humans, like seeing and moving through a room, are the ones that are hardest for machines, while abstract reasoning is comparatively easy. Decades of progress in sensors, compute, and machine learning went into closing the gap between the Cart’s halting crawl and a modern car that drives itself through city traffic.
Why business readers should care: the Stanford Cart is a cautionary tale about demos. A system can clearly prove a concept while remaining hopelessly far from useful, and the distance between the two is usually measured in the unglamorous work of making perception fast and reliable, not in the headline capability.