Why testing self-driving cars in SF is challenging but necessary | Hacker News Reader

Why testing self-driving cars in SF is challenging but necessary

medium.com

20 pointsmdm_3127 comments

There was enough hard data out there back in March to see that General Motors, that creaky old dinosaur of a car company was kicking ass in self-driving, but all my efforts to point that out to HN have been met with grief. Look at what their cars can handle, it's amazing!

I got a laugh when a post about GM's "Super Cruise" got shot down as inferior to Autopilot since it's "Just LKA and ACC but it only works on highways and needs mapping".

Autopilot is actually LKA (w/ LC) and ACC, and AP1 actually used the same Mobileye sensors most cars with LKA use. GM was shipping cars with similar sensor suites years ago but only using them to warn the driver, rather than actively lane center.

Super Cruise on the other hand is the first, hands free, LKA/ACC implementation that's actually designed with more than the basic features current LKA/ACC implementations have.

It's also the first one that feels like a step towards full self driving rather than an OpenCV exercise applied to a real car because of the use of LIDAR mapping data being used to react more like a human would expect (like slowing down before curves and not veering at the top of hills, etc.) and the confidence to allow the user to take their hands off the wheel.

I'm more interested in what Cruise Automation, GM's robotaxi subsidiary is doing. The in-house Super Cruise efforts are roughly on-par with the various autonomous features that are trickling out from a bunch of OEM's. The incremental approach to autonomy is maybe intuitive to Automakers, but you've got to deal with the vigilance decrement, and the handoff problem, and all the complicated matters of policy, liability and regulations surrounding these incomplete systems that still depend on keeping the driver in the loop to one degree or another. What's the real gain to be had from implementing a bunch of gimmicky features in the comparatively small luxury car segment?

Robotaxis on the other hand stand to be disruptive in the Clayton Christensen sense of word, and even if it wasn't, I'm pretty firm in the belief that being bold and taking on the whole self driving problem as a monolithic entity is in the long run more efficient than trying to do it piecemeal.

You dont have to look far, the article is written by Cruise’s CEO. :-)

TL;DR "We choose to test in San Francisco, not because it is easy. But because it is hard"

More like "We choose to test in San Francisco because that is where we are located."

They also are testing in Phoenix. No reason they couldn't do all their testing there.

Reply on news.ycombinator.com