Comment by xyzzy21 | Hacker News Reader

Comment by xyzzy21 | Hacker News Reader

Very true. Waymos are held in considerable contempt in Chandler. Local HATE them (we have a plant in Chandler).

But more broadly, the "learning territory" is bizarrely and truly inapplicable to 99% of the rest of roadways in the USA.

Chandler roads are:

* Unusually flat * Unusually broad * Unusually straight * Unusually well-lit at high sun angles (no shadows) * Unusually good weather most of the time (ignoring Haboob sand storms and monsoon downpours that happen for a few weeks in the summer) * Unusually light traffic (compared to any urban center)

Honestly they are a MINIMUM of 20-50 year away from "any road in the USA" and that still presumes radical and unknown improvements in the technology arriving in a timely fashion!

I'm not holding my breath on this technology! I'm OK if someone wants to blow through some serious cash - that's their choice. But it's not going to pay off any time soon. Fools may believe otherwise but fools and their money are quickly parted and that's on the fools.

> Local HATE them... truly inapplicable to 99% of the rest of roadways in the USA.

Have you considered they might be training the vehicles on you, not the roads? At some point roads become a solved problem and understanding/anticipating other drivers becomes thr hard problem.

Yeah this is definitely most of it. Bad weather certainly doesn't help and can be a problem but people are the main issue.

That's why they are now expanding to San Francisco:

https://arstechnica.com/cars/2021/05/real-robotaxi-service-g...

It's a PR stunt and an attempt to continue to show "progress" so they can keep getting funding. Their current design requires up-to-date ultra high definition 3d mapping, perfect weather conditions, and a minimal amount of traffic weirdness. Maybe 5g gives them a way to keep up to date 3d maps onboard, but even if that works out the data costs and cost to continually re-map the streets makes it so the unit economics don't work. It's not like they just need a few more miles of "learning", the foundation of their entire strategy is flawed and can't scale.

Weather and traffic - fair criticisms.

HD Maps - really aren’t as difficult as you’re making them out to be. Most AV companies can update their maps using sensor data from their own fleet. And the cars need to recharge/service at a facility every few hours anyway. So there certainly isn’t any need for 5g or expensive cell data to make HD maps scale.

It's not in this article but I read it in other on ars that they go to SF because it turns out that they was on too easy roads.

That's encouraging. It's hard, and there's a market. If and when they can eliminate the "safety driver", it's real.

DMV has three levels of license for self-driving vehicles.

- "Testing with a driver". This has about the same restrictions as a learner's permit - must have a licensed driver onboard, no vehicles over 10,000 lbs, no carrying passengers for hire. Over 50 companies have that license. Uber flunked this and had their license to drive autonomously revoked in 2016. So Uber went to Arizona and killed a pedestrian. Then Arizona suspended their license and Uber stopped.

- "Testing without a driver". This is tougher to get. There are specific technical requirements. There are communications requirements. A "law enforcement interaction plan" ("How do we stop this thing?") SAE Level 4 or 5 is required; 3 is not good enough. Operation may be restricted to certain types of roads (freeway, campus) and limited in speed. Eight companies have this license now, which is encouraging.

- "Deployment". Only one company has qualified: Nuro, which makes slow-speed shuttle buses. This is the Commercial Drivers License level for self-driving vehicles.

DMV has a staged approach. It seems to work. Companies can test, and failures get reported and publicized. Complaints about the California DMV being too restrictive about testing have subsided. Most of the companies that couldn't meet the requirements are gone.

There's slow, steady progress. Alphabet/Waymo and GM/Cruise are getting close to deployable systems.

50 years? That's like... the entire history of modern computing.

Why do they hate them?

they're taking our jerbs!

I live and work in and around the Waymo service area. Personally I have never heard anyone express annoyance with the vehicles, so I'm not sure where it's coming from. Driving I've never noticed them to affect me in any way.

Only one example I remember. I was going down a residential street to a midblock through street, usually I wouldn't expect any other cars. I pull up and find my self 6 cars back behind a Waymo to turn. And waited. The Waymo would just not turn whereas all the humans would have. The street has relatively constant traffic.

The only thing I can think of is that people are annoyed by them going the speed limit. Some streets in Chandler are 45 and I swear everyone is trying to go 60, so I suppose people that people that want to be going well over all the time get mad if they are stuck behind a Waymo. In Tempe the traffic tends to seem slower and more controlled so possibly they aren't as bothered by it.

Those conditions are probably why the area was chosen: because it's a good place to start safely, collect data. They are working out the tech probably less so the business model.

It's likely they can't deploy in some areas because snow/rain are just too hard to navigate safely at this point.

The trillion dollar question of course is are they 'almost there' or is this an 90/10 thing where the last 10 is really the 90.

Reply on news.ycombinator.com