As one example, the short train between the Oakland BART station and Oakland airport has no driver. Ever. And no place for one. I think there is some access panel where I suppose someone could plug in controls.
And it certainly was not the first such train.
How deep does the snow get in Mountain View? :rolleyes:
Bolding added.
There are a lot of benefits to self-driving cars - safety, reclaiming your time, etc - but the bolded part isn’t one of them. It just isn’t going to happen. For two reasons: first, while it’s true that automated cars that communicate can travel closer together and make more efficient use of street space, any freer-flowing areas will be immediately consumed by induced demand. More people will drive to work rather than take transit, or take a quick errand for lunch, or to pick something up, etc. It’s like opening a new road, there’s more capacity, but it’s almost immediately used up. But opening a new road basically causes stasis in overall congestion: it’s not better after the new road opens, but it’s not usually worse either. The second reason you won’t clear up traffic problems in urban areas is that the additional road capacity caused by automated vehicles comes with an increased willingness to tolerate traffic. If you’re sleeping, or reading, or watching a movie - or not even in the car - you’re more willing to put up with the horrible traffic. Automated vehicles will cause such an increase in demand for road-space that it will more than compensate for the efficiency-induced increased availability. Traffic will be worse once fully-automated vehicles become wide-spread. Count on it.
I don’t think many people realize how autonomous cars are going to be sold and marketed.
First, it won’t add just $5k to the vehicle cost. Try 10 to 50k. Second, it’s not a one time cost. If you want the autonomous car to go anywhere, all maintenance must be performed. Once the computers or sensors become deprecated, they must be replaced with newer, better computers and sensors. All critical components (brakes, tires, etc) must be inspected by a licensed professional according to a service interval - and the car will track this.
Once the car framework becomes deprecated, it is to be recycled.
Nobody wants to “own” an asset like that, since they won’t really own it. Much better to lease. And since leasing is expensive, much better to rent it out to as many people as possible and just borrow an auto from the pool when you need to go somewhere.
The reasons for all this is because the manufacturer is 100% liable for every crash. So they aren’t going to keep autonomous vehicles on the road with known flaws, or try to maintain a second set of software to run on older computers in older models. Better to just recycle and rebuild the vehicles.
I think the ash from the fire may be piling up.
Actually, they want to test in snow, they drive it to Tahoe. And California buys these things first anyway. I assure you, when I get old, if I’m still living here, the inability to drive in snow won’t matter a damn.
Though of course it will be fixed long before then.
In some areas, mass transit is unfortunately not a significant factor in a commute, so any avoidance of it will not be significant. There is no reason that autonomous vehicles cannot be bound by carpool lane laws, which should cut down traffic.
But the main reason is that a lot of traffic woes come from collisions. Reduce the number of collisions and you reduce the congestion. Another factor is irrational driving by a minority. For instance we have dive bombers, that cross three lanes from the fast lane to an exit in a few hundred feet. Get rid of them and things will improve.
So traffic will be a wash or may even improve.
None of that matters. Drivers switching from transit is only one source of pent-up demand. A lot of trips aren’t taken (or are time-shifted) in metro areas to avoid traffic. Lessen traffic, and those trips will happen. It finds its level. There are only a few places in the US (Kansas City, eg) that could reasonably be called metro areas where infrastructure is sufficiently built out that there is little pent-up demand, and in larger metros, there’s just no opportunity to build new freeways. If you wanted to add new freeway capacity in Seattle, for example, you either have to tunnel, or build it on top of existing freeways. That get pricy fast.
The difference between L4 and L5 vehicles is whether the external conditions are controlled or not.
L4 testing is happening; Navya for example was going to be trying out 2 autonomous shuttles at UMich this fall (though it doesn’t seem to have kicked off yet). I believe there will be L4 vehicles running somewhere in the next 5 years, but they will effectively be trams, i.e. running in very closed circuits.
The L5 example I always give is that I wake up in the morning in Boston, in the mood to see a playoff game at Wrigley. I program my Tesla to do so; on the way there is a snowstorm (of the type I could very carefully manage myself), followed later by heavy freezing rain, followed by plastic bags blowing across the road, followed by a brutally bright sunset in my face. And I would like to sleep through all that.
I think at least 30 years for that to happen and for those cars (Tesla or whatever) to be in the economy at a price that non-millionaires can afford.
My 2 cents.
From reading google’s document on the matter, in the scenario you outline, their vehicle would likely find out about these extreme conditions and refuse to make that drive in the first place. If it found itself caught in these conditions, with no reliable ability to see around, it would do the best it could to bring the vehicle to a safe stop.
You as a human cannot safely navigate in these conditions either. The fact you haven’t died yet doesn’t make you safe or reliable in that kind of weather, it just means you risked your life and got lucky. If your vision can’t see more than a few feet in front of you, you are a danger to yourself and everyone around you if you continue to drive.
If conditions are such that you can see a considerable distance ahead, enough to safely drive, but there is occasional pieces of debris and snow falling, current computer vision techniques will work under those conditions. The pieces of debris can be classified correctly, the car autonomy software isn’t as stupid as you think, it will have seen things like plastic bags thousands of times and have a memory* of what they look like. Snow can be filtered. Also, plastic bags do not reflect radar.
*Even the very earliest autonomous cars have this type of memory. It’s just what you think - a database of snippets of pixels tagged with data about the object.
I didn’t say a blizzard, and no, AV still can’t manage moderate snow falling on a snow-covered road. [you seem to be swinging between blizzard and occasional snow, and that was not my point]
“Plastic bag” is shorthand in the AV world for things that skitter across the highway in a somewhat lifelike way. Here’s an example.
We aren’t close on those.
There are already a few states (e.g. Arizona) that have minimal regulatory hurdles.
I’ve never understood why people are concerned about insurance. Why should this work any different than now? Your insurance policy is, and will be, responsible for the accidents caused by your vehicle (plus the usual coverage for when an uninsured motorist runs into you). Your insurance company will charge you more or less for a self-driving car depending on whether it thinks self-driving cars are safer or less safe than human drivers. If an accident is due to some defect with your car, your insurance company will sue the manufacturer, same as now.
If your car’s computer really does drive better than people do, insurance won’t be any sort of problem at all. Insurance companies will encourage you to have a self-driving car, they’ll offer discounts to get you into them.
Insurance companies will lose a lot of personal liability premium once the L5 cars arrive. And a lot of the remaining legal liability will shift to manufacturers, who will likely retain part of the risk, and then buy product liability insurance to cover the rest.
You might remember that when GPS was new there were many stories about GPS directing people into rivers and stuff, and some people dumb enough to follow it. If snow is that big of a problem (unlikely) the first markets will be places like where I live where it never snows. But driving in snow is not all that difficult, any idiot can do it and based on accidents I saw when I lived in New Jersey, many do.
“If snow is that big of a problem (unlikely)”. Snow is in fact a substantial problem, not just because it obscures the road, but because it gathers on visual sensors; and in New England and other places there’s salt and dirt flying up from the road, which screws with all the sensors.
If you have to restrict the environment, we’re back to L4. That’s not a bad thing - I believe that a place like Las Vegas is ideal for L4 vehicles, either doing circuits on the Strip or running back and forth to a dedicated stop at McCarran. But even that’s a while off.
Snow is a substantial and very surmountable problem. There is no reason to expect that the technology will not be able to handle the issues associated with snow within a few years.
The techniques include pretty much what human drivers do but supplemented with additional channels of information and redundancies.
Like humans cars will be able to use landmarks and both programmed and past knowledge of the road. We know where there is supposed to be a stop sign and behave concordantly even if we cannot see it. We use other landmarks and other vehicles to deduce where the road markings should be. We of course just have our field of vision and what we see in our mirrors when we look as our inputs. The vehicle has a larger field of vision that it is always completely aware of plus other multiple other channels to compare with. Like us it can use algorithms to discount rain and snow as “noise” and not objects. And of course increasingly cars will have V2V communication announcing where they are, where they are going, and what they are even going to do, to other vehicles, which will use that as a major part of the deductive mix as well.
Dirt gathers on the visual sensors? Oh my. Unsolvable! They may need to develop some incomprehensible technology, like placing the sensor behind a bit of clear glass that defrosts and has wipers, like they do for our visual sensors.
Is it all there yet? No.
How far away is it? My guess, and the guess of those who are investing some gadjillions in it, is that they are several years away … not decades, and not never. How long between when the technical problems are solved and they become a large share of new car sales? That’s the harder question.
What’s your opinion on them simply replacing personal car ownership with pooling? There’s obvious a bunch of efficiency advantages to that :
a. The primary cost of a car is the depreciation. If the car is used 30% of the time instead of 5% of the time (that is, 8 hours a day it’s moving someone around instead of 5% of the time), that effectively divides the depreciation into a much larger number of operating hours.
b. The second largest cost of a car is the maintenance. If the car is run for 300-500k miles over 5 years and then recycled, it creates incentive for the buyers of these fleet cars to pick cars that have low per mile maintenance costs.
c. The third largest cost is fuel. Electric is 40% the cost per mile as gas. Pooled cars make EVs more feasible because there’s no such thing as range anxiety. Want to go somewhere 300 miles away? The pooled car that picks you up will be a hybrid.
Also, autonomous EVs could plan their recharging breaks to take advantage of lower time of day electricity prices. (whenever the sun is bright on a cool, windy day, this creates a glut of solar power. Already happening in California and is probably going to happen elsewhere in a couple years)
d. All those costs above would be halved if you car-pool, but it’s very inconvenient to find a car-pool partner at some arbitrary place and time of day and destination. Far easier to find partner matches if it’s done from a vast pool. (and with reputation scores so you can choose to only carpool with people other people like if other people also like you)
e. Many Americans choose a car to handle all their needs. Cheaper cars could show up when they don’t need as much car, specific to immediate needs. Such as 2 wheeled gyro pods for long commutes to work, a minivan when it’s time to pick up the kids and go shopping, just a hatchback when it’s time for mom to shop without the kids, or a luxury car when it’s time for a nice day out. A pickup when you actually want to haul stuff from the store.
Anyways, if (a)-(e) really translate to thousands of dollars saved per average American family per year, at actual pool rental rates, a lot of people would start choosing to do it. Not all at once, and they’d hang on to their current cars, but stop buying new ones and maybe reduce themselves down to one car.
Sure, upper-middle class Americans like doctors can choose to spend more and stay with a personal vehicle, even a personal autonomous car, but the percentage of the population who have enough income that a few thousand bucks a year is insignificant is only a small percent.
The very article you linked to concludes with “So, while self-driving car technology is getting better at handling snow, it’s clear that we are still a long way off. In other words, don’t expect a driverless car to drive you through a blizzard anytime soon.”
Not sure why you’re being sarcastic, but that incomprehensible technology you’re referring to, which has been around for a century, often does an inconsistent job. Also, salt causes refraction for some technologies (can’t recall which ones).
The issues I’m describing are the same ones the OEMs are struggling with, which is why some of them have become more careful. A couple of car manufacturers are still saying L5 by 2020 or 2021, but many are describing L3 or L4 as main focuses for now, and one of them at the last CES said L5 was WAY off for everyone.
SamuelA, I’m not alone in agreeing with you that ride-sharing (along with trucking fleets) will be an early adapter. And clearly there is an increasing population that relies on ride sharing services in preference to ownership. Lyft and Uber have already launched pilot projects. Didi, the big ride-sharing company in China, is also in the space. It’s where the puck seems to be heading.
Maserschmidt, that “inconsistent job” works fine enough for our vision sensors. I’m really not getting why you think that AI with more redundant inputs than we have cannot handle understanding what is a smudge or a smear and how to perceive around it as we do. FWIW I think 2021 is likely on the optimistic side for L5, but conversely more than a decade is hanging crepe.
I remain steadfast in my belief that the biggest challenge is not snow, that’s a very tractable problem, but erratic human driving in other vehicles which may become more aggressive knowing that the autonomous vehicle will “let” them cut them off.