Ars Technica: Experts weigh in on whether lidar is necessary for self-driving cars

New article by Timothy B. Lee, who regularly covers autonomous vehicle technology:

Check out Timothy B. Lee’s replies to tweets about the article as well:

Great. Another article presuming to authoritatively present the conclusion of ‘the experts’ after talking to a small number of biased individuals.

The statement ‘anyone relying on lidar is doomed’ is pretty straightforward and easy to have confidence in. To wit: if you are ‘relying’ on lidar then it means your vision technology isn’t good enough to meet your application needs without it. The function of lidar as implied by it’s advocates is to provide verification of free space around the car, so if you are relying on it then it means that your cameras aren’t good enough to detect free space reliably. But free space detection is table stakes for cameras in self driving applications. If you can’t detect free space then you’re not going to be able to do any of the things that are much harder than detecting free space but which are also critically necessary for self driving vehicles.

Alternately said - any vision system which is good enough to be used as a component of a self driving vehicle is inevitably good enough to do free space detection so well that the only function lidar provides is completely superfluous.

So if you are ‘relying’ on lidar you are indeed doomed to fail at the task of self driving.

What are the harder vision problems than detecting free space?

Gaze detection, pose determination, road hazard classification, occluded space risk evaluation, … The list of required FSD vision capabilities that are harder than free space detection is endless.