Hacker News Reader

Writing software for competent self driving is turning out to be really hard, and no one is succeeding. On the other hand, building better and cheaper lidar hardware is a well-defined problem that companies can pour money into. Their software still sucks, but everyone’s occupied with wrangling the $100k lidar.

View on HN

I agree... but I also want to point out that it's totally possible that writing software to make self-driving cars work acceptably well might be easier if the software has better sensors than a human does, so it might actually be the best way to try to solve the problem.

Could some rebel make things like fake road bumps that lidars can't pick up, but humans can?

Protecting AI against spoofing attacks is a pretty active area of research right now.

Protecting neural networks from adversarial attacks, i think you mean. Protecting good A.I. from spoofing attacks is exactly the same as protecting human drivers from spoofing attacks.

Or the other way around - trick the lidar into thinking a road bump is there when there is no bump.

Reply on news.ycombinator.com