I think that you don't understand snow (real winter) and how it effects vehicles. A dusting of snow that stays on the ground for a day before melting is not an issue. The issue is when the snow keeps the road markings covered for literally weeks or months at a time. When the lanes are not visible to human drivers and they form new flocking based emergent lanes which all humans follow instead of the actual lanes. When the snow piles on the edge of the road change the width from week to week and force parked cars out into the middle of the old lane.
Snow is not a problem. Snow that stays is a problem. Atlanta doesn't get snow that stays. Waymo is noticbly absent from Buffalo after their one prior attempt.
> you don't understand snow and how it effects vehicles
Tell me more about how the 92” of snow my town got last winter leaves me ignorant.
> Snow is not a problem. Snow that stays is a problem
Snow used to be a problem! It isn’t anymore because it’s solved. My Subaru can keep lane using radar alone, following the car in front of me, in a blizzard.
> Waymo is noticbly absent from Buffalo after their one prior attempt
They’re also noticeably absent from Chula Vista [1].
Also, I know I don’t understand snow, but maybe the folks in Denver do [2].
> want to know where I should be steering clear of Subarus in the winter ;)
We’re actually pretty good! The fuckwits are in the FWD rental cars that can’t brake, ever, and souped-up F-million fifties driven by rich 17-year olds who predictably flip them on flat straightaways despite infinite farmland run-off, at grade, on both sides.
And to be clear, I’m never leaving the Subaru alone. The Subaru isn’t letting me leave it alone. But the notion that Waymo couldn’t figure out snowstorms is one I’ll readily challenge given the Subaru’s radar frequently sees white cars in a white out before I make them out visually (at 15 mph with hazards on). In the snow, an autonomous vehicle’s radar (note: not lidar and certainly not cameras) have an advantage over humans.
When there are no lane markers, the two scenarios are there's a car to follow, or there isn't. If there a car or cars to follow, that's easy enough, just follow behind at a safe following distance * it's
snowing factor. The other case that people seem to think is impossibly hard, is for the computer to invent lanes because now there are no lines for it to look at. It's harder than if there are lines to follow, sure. And it depends on the type of road you're on. A busy city street isn't the freeway, isn't a suburban stroad, isn't a windy rural road. But fundamentally they're the same. It needs to "know" the dimensions of whats drivable (including if traffic is one or two way), figure out how many car widths will fit, including an error margin so other drivers get a bit of extra room, and then do basic math to divide that into lanes, and then just pick one.
A pre-AI regular computer vision algorithm could do that. Combined with the fact that Waymo maps out locations in high detail before offering rides in an area (which means the edges of what's drivable is already known to the AI, and, I mean, they don't drive there currently, so me pointing out it's basic math isn't, like,
proof or anything, but it just doesn't seem insurmountable, given the other things computers can do these days. Computers can look at photos, tell you what it's a photo of, and then you can search for "car" and show you all of your pictures of cars! OCR works well enough that I can take a photo with text in it, and then just copy and paste the words, without having to wait for the computer to run a slow analyze process that I have to wait on first.
Computers are fallible, but the other part of that is, having driven in snowed out streets, sometimes you get stuck behind someone who's snow covered lane math you don't agree with, and you either pass them or get stuck behind them. Which is annoying, sure, but it's one of those things you just kind of live with during winter? Complain about to your friends and family maybe?
The other thing is that we all know that in dry sunny conditions, city streets aren't always well marked. So there's already capacity to put the vehicle somewhere that's not rigorously defined by painted lines on the street. Now,
we don't know how much Waymo has their army of contractors manually add lines to the street data that the cars have, or if the computer calulates that, which would make it harder to drive when there are no visible lines because it's been snowed over.
Anyway, the other challenge about winter driving is the difference in traction. There's snow, ice,
black ice, wet asphalt, salty asphalt, and sometimes even dry asphalt. 4wd is popular on cars in those areas to help deal with varying conditions. With 4wd, moreso AWD, and also ABS braking, we're already relying on a computer to sense how much the wheels are slipping, and then to transfer power to the wheels that have traction. I'd imagine a more advanced computer could help out those systems and preemptively tell them what's going on. (Black ice btw is why I don't believe in camera-only systems. It's not that cameras aren't capable in regular conditions, its that I think self-driving should be better at things, and if it's got data on where there's black ice because its advanced sensors just simply pick it up, we'll all be safer.)
Only time will tell. I'm just a random on the Internet who took some computer vision classes while I was getting my. degree back in the OpenCV days. Maybe it is that hard and self-driving cars never make it to snow prone areas. I just don't think it's as big a challenge as some people make it out to be. A system with LIDAR should be able to gauge distances in the dark better than I can with my human eyes that need headlights in order to make out anything after the sun goes down. Which in the dead of winter, the sun goes down at 3pm, which really messes with the human psyche.
When were you last in a Waymo? I use them almost exclusively in Phoenix.
> they stick to the cheat regions
Do you think it doesn’t snow in Atlanta?