Self-driving cars: Will they snow?

"Old-Busted-Hotness" (old-busted-hotness) 02/05/2014 at 10:09 • Filed to: None

0

8

Just wondering. Since we have 5 inches overnight and our street won't be plowed until May, I'm well acquainted with driving on this slippery shit (including Scandi-flicks and handbrake turns). It's something you get used to doing when you grow up in Ohio and are too stupid to leave.

What about self-drivers, though? Do they know enough to apply wheelspin to get forward progress? Or will they just sit there and lurch as the traction control kicks in repeatedly? Or do they go full HAL and say "I'm sorry Dave, it's not safe to go out. My weather report says the roads may be icy."

Have the self-driving cars even been tested in the snow? What happens when the sensors are covered in a layer of ice? And who do I sue when I crash one?

Casper > Old-Busted-Hotness 02/05/2014 at 10:11

0

They should be able to handle snow better than people. Self driving cars are already doing everything based on surface friction, instrumentation, and laser/radar/IR scanning. So technically they can see and feel more detail than a human driver. It will just depend if the little caffeine addict that programmed the algorithms did his job.

If you crash one because you failed to maintain it, it's still just like you failed to clean your own windshield or maintain your tires... and that means it's on you.

deekster_caddy > Casper 02/05/2014 at 10:18

0

I don't know, I'd like to see a self-driver go through the 4-6" of loosely packed powder with no clear lane markings that I sloughed through this morning. Would it give up on the 'official lanes' and follow the existing ruts in the snow? Or try to mimic what other cars around it are doing? I would expect a self driver to end up going about 5 mph on this stuff, then not having enough momentum to get up the next slope.

jariten1781 > Old-Busted-Hotness 02/05/2014 at 10:22

0

I'm guessing that in extremis the cars will default to manual operation. They'll have to be able to handle plowed still-slick roads before they get adopted anywhere, but I doubt that they'll be programmed for every corner case.

Casper > deekster_caddy 02/05/2014 at 10:22

0

Sensors can see through snow if they don't operate on visible wave lengths many times, and still see lines. As far as traction to clear a hill goes, that would be the hardest part, but most likely in many cases the systems would be programmed to take a longer, less unreliable route. If they do get stuck, it's up to the driver to get them unstuck just like any other car.

Why would you expect a self driver to go slower than a real driver? Self driving cars would in theory be able to do everything faster than a real person. It would all depend on the processing power and what sensors the systems had access to.

deekster_caddy > Casper 02/05/2014 at 10:35

0

If it's based on traction/stability control feedback, it's going to end up slowing down when driving in heavy snow. Technically all of us going 25-35 on the 4-lane highway today were driving with very little traction most of the time.

One legitimate (I think) concern is drivers getting out of practice. The self-driver handles 98% of the days, then you get a day like today and it's somebody's first chance to drive all year. Driving once a year does not keep you prepared for driving in really crappy conditions.

Casper > deekster_caddy 02/05/2014 at 10:40

1

That all depends on the programming. You only need the traction you need for your next few probable tasks. For instance, going down a highway, you only need the amount of traction you need to stop based on the distance to the next vehicle to be "safe". It's completely possible that developers could decide the margin for safety isn't 100% braking distance, but more reasonably 50% or 30%. The car is looking forward and calculating distances, paths, etc, so it can be predicting many of the problems very accurately. Then you can also consider that actual traction is only a small part of the equation for avoiding an accident. To completely stop, you need significant traction, to avoid an accident you only need a portion of the potential traction, and escape vectors that you have calculated to use. Think of it in terms of motorcycles, rarely is the plan on a motorcycle ever to stop moving.

I think we have proof of drivers being out of practice with all the weather that has happened lately. People don't consider driving a necessary skill or think "driving" can be encompassed by putting a car in gear and going from point A to point B. I don't know how much worse they can get.

Old-Busted-Hotness > Casper 02/05/2014 at 11:40

0

I dunno. I know I can see a hill up ahead and know that I'll need a bit of speed to clear it. I know how to keep control when the drive wheels are slipping, and my butt gives enough feedback to know if I'm gaining or losing speed. The first part there is predictive, the latter is reactive. How predictive is the software? I'm assuming the cars have maps and GPS so they know where they are, and GPS does a pretty fair job with altitude ASL, so they know when there's a hill coming up. Maybe, though, they simply react to conditions.

Snow would be one hell of a test for self-drivers.

As for the sensors, can they "see" through the dirty slush that accumulates on your bumper and grille? You know, that dark gray mixture of road salt, spray, dirt and ice? Anything radar-based should be OK, but sensors that use light (visible or not) would have a real problem with it.

Casper > Old-Busted-Hotness 02/05/2014 at 11:42

0

Everything you know a computer can know, everything you can see, a computer can see. It all depends on the components and what processing logic we give it. Maintaining the system is still on the driver. Many sensors used by self driving cars are on the roof, not the front bumper, and some are even mounted inside the windshield so the wipers can double duty for the driver and visual sensors.