Hacker News Reader

Comment by someonehere | Hacker News Reader

They definitely aren’t ready and should move away from operating in SF altogether. They should operate out larger cities with wider roads or limit where they operate in SF like bus lines.

I think this compromise needs to happen because SF didn’t plan for robotaxis when the roads were built decades ago. Towns like Phoenix and Las Vegas have the roads to support it. So they should consider splitting operations that way to continue working.

This does nothing but make other towns think the taxis aren’t ready and will fight to keep them out of their towns.

It’s an optics issue at this point if they want to convince people the robots can drive better than humans

View on HN

I think its certainly a matter of when, not if robo-cars are capable of driving as good as/better than human. Sooner, rather than later too. That said, totally agree.

SF seems like the worst place to pilot this, for so many reasons.

Definitely a matter of when, but that’s also the case for nuclear fusion. Trouble is these perennial 5-10 year away things seem to always be later than we expect.

It’s probably the best from their training standpoint. The city is quite small relative to others and their only hope is to essentially memorize San Francisco in the neural networks. IMO you could not take a cruise and drop it into any other city and have it work.

what stopped them from memorizing a different small city in the US? there are tens of thousands to choose from.

it seems to me like their decision to start with SF instead of something a hundred times smaller was hubris

No, test your silicon valley death machines among yourself, please.

Are current cars driven by humans also death machines?

People can be held accuntable. Robocars have a more murky record.

But you still think that anytime you drive you make the decision to sit in a murder machine and no matter what the reason is any accident is your fault and you should go to prison? even if the breaks fail due etc

I can't speak to Cruise, but I strongly believe Musk set Tesla back a decade or more with his demand to use only video cameras.

We have lidar, infrared lasers, polarized ultraviolent imaging, multi-microphone 3D audio sensing, triangulating RF antennas. Why limit yourself to the limits of human perception?

I’ve heard that lidar is somewhat unfavorable because it’s so susceptible to becoming useless due to misalignments from something as simple as a pothole. Radar and other options though are absolutely worth using, I think musk is just trying to cut costs. Their S/X are using radar with the new FSD HW4

What makes you believe that? None of the research supports "as good as" being anywhere on the horizon, let alone "better than".

As an aside: Why are robocars practicing on wide, sunny roads in the American southwest? Shouldn’t they test on the hard cases? Rain, Snow, alleyways, streets, that exist since medieval or earlier times, traffic which takes a laissez-faire approach to rules?

They’re too deep, their networks are overfit to San Francisco at this point. Making it work in other cities would require the insane training hours to basically memorize that city.

That’s not how it works at all

Nobody is training ml models to “memorize the city”. They use cm-precision maps

Cruise has been operating in Phoenix for many years now.

Should have gone to Dallas and its surrounding suburbs.

Reply on news.ycombinator.com