Hacker News Reader

Comment by fred_is_fred | Hacker News Reader

Has the issue that self-driving cars have with snow been solved? Half the country would like to know.

I see this said frequently about self driving cars and I'm just baffled by the cynicism. Who cares if self driving cars can't drive in all conditions yet? A shit ton of people live in parts of America that are hardly ever subjected to poor weather/snow, perfect places for self driving cars to start in. And in these early stages, if I tried to get my car to self drive in terrible rain/wind/snow and it just said "can't do that sorry" I'd be fine with that.

Your attitude is like looking at the wright brother's plane and being upset it can't make a transatlantic flight yet.

Give it some time

I'd buy a self driving car that just doesn't work in bad conditions.

The bigger question is "has the self driving issue been solved in clear conditions." Is there any real evidence someone has a self driving car that is better than human drivers?

I see a lot of promises.

Oh, no. GM doesn't expect you'll be able to buy your own for 10-15 years. They're focused on robotaxis.

Robotaxis are geo-fenced. They're worth ~$200K apiece (that's a very loose figure). Autonomous vehicles will require a fair bit of maintenance and oversight before they're idiot proof enough put in the hands of private owners. Initial commercial deployments will certainly be in fair weather cities.

If GM can make more money operating a robotaxi fleet, why would they ever even offer them for sale?

There are a lot of medium sized cities in the US with crappy public transportation systems. Maybe these cities could subsidize on-demand rides in robotaxi fleets operated by GM and Uber and other companies rather than owning and operating a fleet of buses running on a limited schedule.

Centennial, Colorado, is already testing subsidized rides (using Lyft) just as you've mentioned, only with human drivers.

https://www.citylab.com/transportation/2017/11/what-public-t...

> If GM can make more money operating a robotaxi fleet, why would they ever even offer them for sale?

Because they can make even more money doing both. I don't think the independent must-own types will magically change their ways, and taxis don't solve many people's needs. Why not sell em too? It's like asking if they make more money leasing why do they sell also.

It's a good question. GM is projecting they'll be able to get rideshare costs down below $1 per mile around 2025. They believe that when they can do that they'll be able to collect 20-30% margins through a scaled robotaxi service. It would be a waste to sell a vehicle at a 5% margin when it's worth so much more as an autonomous fleet vehicle.

How do you get to 20-30% margins off of $1/mi costs? Uber and Lyft are already <$1/mi for pooled services and around $1-1.50/mi for non-pooled services in low cost markets. Price competition should make anything beyond classic transportation margins of (negative) - ~5% standard for any self-driving taxi service. 20-30% gross margins might be standard but operating margins at that level seem like a pipe dream.

They also mentioned data monetization as an unspecified portion of those 20-30% margins. Robotaxis will gather a lot of data, around 4 terabytes a day. Enough to hypothetically run a real time virtual simulation of a city, if 1000s of sensor riddled robots are all prowling the streets, seeing, recognizing and classifying everything. I can think of all kinds of ways to monetize data like that, both good and terrible.

Check out their investor presentation which compares their idea of AV ride-sharing with existing ride sharing services:

https://www.gm.com/content/dam/gm/events/docs/5265893-685163...

The robotaxi fleets only need so many cars, and they could always start off selling vehicles at a larger margin to those that really want them.

They're throwing away money if they never sell the cars.

Because other will sell autonomous cars. And the first on the market will certainly have a big advantage.

Also, would you rather take an autonomous cab in a car you own and trust (or that your friends own), or in a car you've never tried ?

I wonder if we'll ever own autonomous cars, or if we'll buy a license to use them like we do with software.

Because selling has better cash-flow properties than renting, and by selling GM can push some financing requirements into smaller and more dispersed entities.

Besides, a bigger market does not mean it's more lucrative. Competition on ride-sharing will probably be much more intense than on car manufacturing, and GM may be completely unable to succeed on that new market.

Automakers sure do make a lot of leases though.

(re your comparison of selling and renting)

As I understand it, from an automaker's perspective a lease is a sale. They're just selling it to a third party instead of the leasee.

Yes, they do. If they do sell the cars, it will probably be for the reasons I pointed.

I'm not saying they will sell them. But the ROI one is a likely possibility.

The bigger question is "has the self driving issue been solved in clear conditions."

It's hilarious how far ahead of ourselves we've gotten on this topic. "Can it drive in the snow yet?" Umm, how about we get them to drive from Redmond->Seattle in clear conditions and minimal traffic before we start talking about whether it'll get me to the ski resort.

> I'd buy a self driving car that just doesn't work in bad conditions.

Same here. I live in the midwest and our two-car family always owns at least one 4x4/AWD vehicle for the winter. The other is usually optimized for MPG/reliability (things that go down when you start adding heavy parts - like the kind needed for AWD).

And I can just drive it myself presumably. I drove a '89 Ford Tarus with bald tires in Chicago winters.

I'm trying to get to roughly the same answer here

https://news.ycombinator.com/item?id=15824953

A lot of it is software. The hardware is mostly solved except for cost and there's no reason to think it won't go through the same cost reduction curves that are common throughout the industry. So, IF you can get to viable technology for specific use cases, there's not much reason to think you can't get to workable economics.

> So, IF you can get to viable technology for specific use cases, there's not much reason to think you can't get to workable economics.

This seems like a good analysis, but are there particularly good reasons to believe we will be able to produce a car that drives through an urban environment, at any cost?

Some day. At this point, IMO it would be foolish to say never or not for a century. We could find ourselves in a place where the current path that's mostly deep learning gets close but not close enough. But we seem to be moving forward pretty well even if the last 5 to 10% takes a couple decades longer than many people think.

But I'm still inclined to think that generalized door-to-door autonomy without a competent human present is still a long way out.

Waymo cars already drive in Phoenix, Arizona (i.e. urban environment) without human drivers.

https://www.wired.com/story/waymo-google-arizona-phoenix-dri...

There is a wide range of ability when it comes to driving cars, these don't have to be better than everyone to be useful, just better than the worst drivers, which I think it's clear they already are.

Plus they learn as a fleet. So when they start out as barely better than the worst legal drivers and start improving then a large set of vehicles will improve every time they roll out improvements to handle some edge cases.

If all the car reviewers get their hands on it and say it's like a timid person lost in downtown Baltimore driving a uhaul truck it's gonna crash and burn

You have to do something well to redeem that or sales will be a flop. It doesn't look like self driving cars will be much cheaper, better handling, cheaper to own or more luxurious than normal car so the only thing left is for them to drive as better than most people.

The kind of review you want is "this thing costs $10k more than a Prius but being able to sleep off a hangover on your way to work is worth every bit of that $10k"

Self-driving cars will first be deployed in fleets that you can request, where the cost of the cars will be amortized over the many people who get to make use of them, making them far cheaper than owning, operating and parking the car. Much of this margin will probably be captured by the self-driving car companies for the near future, but in the long term with a competitive market they should be far cheaper than current cars for the vast majority of users.

1) Why do you think that? Has any car ever been successfully test in traffic for long periods of time without human intervention.

2) Better than the worst isn't good enough because good drivers would buy them too.

Cruise Automation reported 6 disengagements over 2200 miles of downtown SF driving last November: https://www.dmv.ca.gov/portal/wcm/connect/deade5b7-5b10-4b25...

Waymo was reporting 0.2 disengagements per 1000 miles last year, but people like to nit pick those because they're not all downtown miles.

For context humans have 0.006 accidents per 1000 miles driven, and that's counting all the simple miles we drive on highways. So last year Waymo was within a factor of 30 of average human performance, assuming each disengagement was an accident.

If you allow for the fact that with so few disengagements, some of this is noise, and that most of these disengagements would probably not result in accidents, but are rather drivers being cautious and the cars are probably better a year later, I'd say they're probably knocking on the door of human safety at least.

Obviously hard to really say without more data, but the data makes me bullish rather than bearish.

A recent study by University of British Columbia researchers found that rain and darkness like one would experience during Vancouver and Seattle winters would also be challenging for self driving cars.

http://vancouversun.com/news/local-news/vancouvers-rainy-wea...

I don't think so, but Waymo started testing in Michigan. https://techcrunch.com/2017/10/26/waymo-to-start-testing-its...

Reply on news.ycombinator.com