vision-only

Vision-only means judging what's around you using only camera footage, without mounting separate distance sensors like lidar or radar. It's the idea that a machine should drive the way a person does with just their eyes.

Tesla publicly chose this path in 2019. Elon Musk flatly declared that lidar was a fool's errand, and most people took that as a comment about cost. But Andrej Karpathy, who stood on the same stage, gave a different reason. Roads, he pointed out, were built from the ground up to be read by human eyes — signs, lane markings, traffic lights, even hand signals at construction sites are all information meant to be seen. Lidar can tell you something is there, but it can't read what's written on it. So whether or not you add lidar, you still have to solve the problem of seeing — which meant, in his view, pouring all your resources into that one capability instead.

In 2021, Tesla removed radar too. The reasoning was notable: mixing two sensors that produce conflicting readings doesn't necessarily raise accuracy — it can increase confusion instead. Reportedly, failures kept recurring where the car mistook a stopped overpass for a stopped vehicle and slammed the brakes.

One thing worth noting: vision-only isn't a way to build things cheaply — it's choosing the harder path. Whatever a sensor would have handled, the vision AI now has to do instead.

"Matic made the same vision-first bet as Tesla" — meaning that even in cleaning robots, distance sensors were dropped in favor of putting everything on cameras.