[pricecomstock]

Hey I made that comment! I assume Waymo has thought about it. I assume lots of people at Waymo have spent plenty of time in NYC. I also don't say that this was impossible or I think it will fail.

What I do believe is that there are tons of edge cases, and diminishing returns with trying to solve for all of them. And, I would guess that it will take quite a while (the decade I mentioned) before self driving technology is capable of making >95% of trips autonomously without extra issues that most human drivers would handle.

Additionally, I think there are moral and comfort issues with AI that need to be addressed. This is speculation, but Waymo likely aims for the appearance of safety and tunes things to be overly cautious. That is at odds with some of the dynamics of driving in NYC, where you need to be assertive and maybe even risky, and I'm curious to see how that will play out.

I can see how it sounds like I'm just shitting on this tech. My intention was more to describe that self-driving in NYC is maybe even a harder problem than it sounds, and possibly an overreaction to the hype about this technology. I think about it pretty often when I'm driving or biking and encounter situations that seem very difficult for a computer to navigate.

[ra7]

You make good points. I don't disagree that there are tons of edge cases and it could take a while. I mostly disagreed with the examples of edge cases that you think makes it hard. Quite a few of them can be solved with an up-to-date HD map (missing lane marking, construction zones, 1-way streets) and there is evidence of Waymo handling pedestrians and other complex scenarios well.

NYC is definitely a harder problem. But Waymo's progress so far has been impressive and they have really solid technology from what I have followed all these years, so I'm pretty confident they can reasonably crack NYC. Your comment just sounded a bit too pessimistic for me is all :)

[StanislavPetrov]

>Quite a few of them can be solved with an up-to-date HD map (missing lane marking, construction zones, 1-way streets)

How do you keep these maps up to date in NYC when lane markings, construction zones, traffic flow, construction and a variety of other things are constantly changing on a daily basis? How do you deal with the ubiquitous "traffic cops" who stand in the center of many intersections and arbitrarily wave traffic in different directions?

[ra7]

Waymo claims their cars can detect environment changes and share it with the rest of the fleet in real time, with most of the map update process automated. They've written a blog post on this topic: https://blog.waymo.com/2020/09/the-waymo-driver-handbook-map...

As for traffic cops, here's a video of Waymo obeying hand signals: https://www.youtube.com/watch?v=OopTOjnD3qY. No reason this couldn't be replicated in NYC as well with some improvements.

[StanislavPetrov]

>As for traffic cops, here's a video of Waymo obeying hand signals: https://www.youtube.com/watch?v=OopTOjnD3qY. No reason this couldn't be replicated in NYC as well with some improvements.

This is an incredibly idealized version of "traffic cop" that doesn't exist in NYC. Here's an extremely well disciplined example of a traffic cop in NYC. They don't stand in one place, they are constantly challenged by traffic from all direction, they sometimes stop certain lanes, and tell other ones to go, with all sorts of gestures. And this is only a 4 way intersection without much traffic, and without any construction or significant pedestrian traffic. Its one thing to program an AI car that can drive under idealized, predictable conditions and entirely another to drive in the chaos of the real world, especially NYC.

https://www.youtube.com/watch?v=m2Z7jS3Yb5Q

[ra7]

These self driving companies are actively trying to solve these difficult problems, so it's not impossible just because it's NYC. If they can go from zero to driving in SF, they can solve the SF to NYC complexity.

[ximeng]

https://m.youtube.com/watch?v=68n4RjGhvRw Has Tesla handling some construction. Not quite there but in theory these cars should handle these situations much like people do - have an expectation of the road layout and adjust based on what they see in reality.

[StanislavPetrov]

I'm not trying to be condescending, but I think there is a disconnect between the chaos of NYC streets and footage like what is in this video. How does an AI car that is programmed to maintain its lane, operate safely and obey traffic rules operate in an environment where there are no lanes or traffic rules? People in NYC zoom around and honk the horn and cut each other off without regard to traffic rules (aside from the red light and speed cameras). With bikers and mopeds regularly speeding the wrong way down one way streets and weaving in between traffic and slow moving/stopped cars. It just seems like an impossibility to me as someone who has driven in NYC for decades. I will tip my cap to anyone who can ever design an AI vehicle that can safely and efficiently navigate NYC traffic.

[rsj_hn]

> How does an AI car that is programmed to maintain its lane, operate safely and obey traffic rules operate in an environment where there are no lanes or traffic rules? People in NYC zoom around and honk the horn and cut each other off without regard to traffic rules

These are all good points, which is why I want to see this test happen. Maybe Waymo can handle it, maybe not -- but this is very useful info to measure the state of the art of a market leader (which I assume Waymo to be).

[StanislavPetrov]

It will be interesting that's for sure.