I am just wondering how many people have to die before it finally dawns on on us that the current approach is idiotic. The supposed smart cars can't even talk to each-other directly 'without internet' to warn of a car pileup! They can't report to city traffic control about their condition. We should be innovating in infrastructure instead - create standardised computer readable infrared road markings, equip each traf…
This seems to be the dominant opinion is here, that is, to approach this problem as we approached civil aviation and flying commercial planes. To respond in a sentence: first, you have no proof that "investing in infrastructure" would actually yield working autonomous cars. Second, there are no proof either that the current approaches are doomed to fail. To respond in details: everything that you propose doesn't seem…
Self driving car has failed to detect a person in the dark and killed her. It was demonstrated that Tesla autopilot fails to detect a child crossing the road. It confused a semi trailer and an underpass. You call that an edge case?
> The response cannot be "let's add a ton of fragile electronics in all of the infrastructure"
So instead we will add a ton of fragile electronics in every car. Instead of having 3 optimally positioned lidars monitoring an intersection, we will have 20 poorly positioned lidars on every car.
You don't need electronis, you can use paint. You can put QR codes that tell a car where it is in an underground parking lot. If we add a parrern to all clothing in infrared-only paint, that would help various AI detect people and tell them apart from surroundings. You can put a metal patterns on street signs that are detected by automotive radar.
> will cost a ton of money to build and maintain
And electronics in the cars will cost even more
> will break in weird ways over time and lead to cascading failures
You mean like the failure the main article is discussing right now?
> it's dubious that it would even work at all
And it's dubious this 'individual' approach will ever work. Every single argument you've made equally applies to the approach you are advocating.