3 ms·

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

by hcks 4y ago

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 to be able to solve the main challenges that self-driving cars are facing. These challenges are all about edge-cases, and responding safely to changing conditions and misbehaviors from other agents. The response cannot be "let's add a ton of fragile electronics in all of the infrastructure" because this: will cost a ton of money to build and maintain, will break in weird ways over time and lead to cascading failures, and most importantly: it's dubious that it would even work at all. The final comment about VCs is weird: a lot of actors in this field are not VC backed startups. My opinion is: Tesla might have a working model that outperform human drivers in 10 years (50% confidence)

ClumsyPilot 4y ago
> These challenges are all about edge-cases, and responding safely to changing conditions and misbehaviors from other agents... The response cannot be "let's add a ton of fragile electronics" 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.

by hcks 4y ago

ClumsyPilot 4y ago

> These challenges are all about edge-cases, and responding safely to changing conditions and misbehaviors from other agents... The response cannot be "let's add a ton of fragile electronics" 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.