I think it goes back to ‘SD cars for what?’. I can see it fairly soon becoming an option on regular personal cars, that doesn’t cost a lot, that can drive the car a lot of the time. There will still be various wrinkles, like the owner’s manual will surely say, to help hold off lawsuits, that you should still stay alert at all times ready to take over. But everyone will
a) recognize that that would make the system of doubtful value if you really had to do that
b) come to recognize that the car is fine driving itself in favorable driving conditions, then you can read (if you don’t get car sick reading, a good point made above) or go to sleep, in favorable conditions (Interstates in good weather, etc).
Then there will still be wrinkles as society deals with the fact that people turn on the feature and stop paying attention and that causes accidents sometimes, including some not very bright people who do it in conditions the car really can’t handle, but maybe distinctly less often than human drivers would cause accidents. IOW a broader version of where things are now, with some more years to avoid some of the shocking lack of capability shown in recent accidents (the Tesla that couldn’t see a truck across the roadway, etc). It will get better than that pretty soon I guess.
But some people mean SD capability undermining the basic concept of personal cars, like people don’t even learn to drive but just get the SD car from the car service via the app. In that case they’d have to solve every problem such as you outline and that might take much longer, like decades. That’s assuming a large % of people were on board with not owning cars or letting them always drive even if the cars could drive in snow so deep you can’t see the road. A large % of people might not be. There’s also the political aspect of ‘saving jobs’ of people who drive for a living. Socio-political reaction to technology is probably harder to predict that technological advancement itself, which isn’t easy.
My Wife has a 2019 Subaru Ascent with all the bells and whistles. Now it is not a ‘self driving’ car, but has lane departure warnings, some sort of assist to keep you in your lane and some sort of dynamic cruise control. It also has traction control that my wife turns off to get up our driveway.
The dynamic cruise is nice enough, the other stuff I don’t really care for one way or another. I’m buying a 2019 Toyota 4-Runner on Monday, and it has none of the ‘self driving/warning stuff’. It does have some sort of traction control that can be controlled by the driver, and a locking rear differential (I want that very much). What I wish they came with is good snow tires from the get go. I’ll have to replace the ‘All Seasons’ I suspect. I’ll test them. I’ve had to replace them with snow tires on every other car I’ve bought. All season tires suck.
Yep. And I don’t see this as a benefit. To be a good driver you have to anticipate what others MIGHT do. Not react to them. A SDC in the above scenario is just another variable thrown into the mix.
I would constantly worry whether or not the self driving car saw a potential hazard or not. It would stress the shit out of me. I may as well drive myself.
In a world of all self driving cars that communicated with each other constantly, this may not be as much of a worry. We are a long way off from that.
Rarely, but sure, those are things that human drivers struggle with as well. A friend of mine died about 15 years ago due to black ice. He was not in a self driving car.
Why does the ranging device have to be on top of the hood. Why wouldn’t it be built into the grill or the front of the car? In any case, yes, a ranging device can range out pretty accurately for dozens or even hundreds of yards, and would easily be able to tell that the road does not match the road it expects.
What it does there depends on a number of factors, but worst case scenario is that it comes to a stop and asks for human input.
Largely the same way that you maintain a straight line on an obscured road. You got GPS, differential GPS, along with other navigation systems that are coming out that promise even higher accuracies. You have dead reckoning navigation, you have object recognition. The car is going to know where the road should be, and so would not have much more difficulty in staying on the road than a human driver.
Once again, worst case scenario, the car stops and asks for human input.
And why would a SD car not do pretty much exactly the same?
And, once again, if conditions are beyond the SD, then you take over. When cruise control was new thing, people were rear ending people and running through traffic lights, and doing all manner of stupid havoc because they were trying to use ti for things that it was not meant to be used for. If people insist on driving an SD in conditions where human drivers are advised to stay home, then it’s not really the SD’s fault if it can’t handle it.
I mean, really, the cars I see piled up in the medians every time it snows indicates that cars themselves are unreliable in adverse conditions.
As you said, it’s the 80/20 rule. You are insisting that something that would benefit people 80% of the time should not be used because it has some issues with the rarely encountered stuff.
The traction control is better than a human driver at preventing skids and other out of control maneuvers. Traction control is very bad at getting out of being stuck in snow. The very nature of it prevents you from spinning your tires, which is what you need to do to get out of snow.
And on that day, you switch to manual, or you switch to your non SD car. Question, can my Ford Focus handle driving up your driveway in the snow? Probably not. Is that an argument for not having economy cars?
I get that you have to anticipate others, I personally drive as though everyone on the road actually is actively trying to kill me, and will do so if I let down my guard for even a second. One of the reasons that you have to react to what people might do, is because it takes you about a half a second to react, once you see that something is happening. That’s at least 40 feet at highway speeds. A computer can react much, much faster.
I personally like the idea of an SD with “passenger assist.”
Have you ever been a passenger, and the person driving did not seem to see an obstacle, so you say “Hey!”, or “Watch out!”, or even just clear your throat?
SD’s should be built so that they can take that sort of input from the passenger, and increase their diligence. Depending on conditions and settings, a vocal warning like that would cause the car to start slowing (not braking, but just “letting the foot off the gas”), it would re-evaluate its input, increasing the threshold for discarding false positives.
But, seriously, do you feel that stressed out anytime as a passenger? You don’t know if the driver saw that potential hazard or not.
No, but I’ve been multiple times in the situation of wanting to throw under a bus a passenger who wouldn’t shut the fuck up or would say the most inappropriate things. I can’t fucking “look at the pretty bird!”, I’m driving!
Huh? Did you miss-write this? No, you don’t want to spin your tires to get out of snow.
The three traction control systems I’ve used operate in two ways. One is to apply braking to a spinning tire. That’s OK, but if your going in a straight line, just basic locking differential is better. The other way it handles this is if it detects even the slightest spin on any wheel, it also decreases engine power. This is what will screw you up. That decrease in power for just a slight spin can (and often does for me) make you loose the momentum you needed to get through. This is the main reason I have to turn it off.
Of course it’s not a reason for not having economy cars.
Understand that I’m playing devils advocate here just a little bit. SDC’s will be fine for most people in most conditions.
I mentioned up thread that I am about to purchase a 4-Runner. It’s about the only car that’s going to work for myself. I have heard that in the not so distant future Toyota may not sell the car in America any more because it is not worth it to them to put all the driver assist gadgets on it. That doesn’t bode well for me.
It needs to be angled downward to see a pothole in the road or the fact that the roadway is washed out under the floodwater that has now receded to only 1" deep. If it’s in the bumper it’s not going to have sufficient time to stop or drive around the obstruction. Pointing it out will pickup something on the road surface for hundreds of yards but not a hole in the surface of the roadway.
Except in every car now, you have a (theoretically) competent driver, at least one who passed a driving test & has a license. In an SDC you might have: [ul]
[li]A child going to/from school or sport practice/game.[/li][li]A millennial who never got a license & doesn’t really know how to handle a car.[/li][li]Someone who went out drinking for the night because it’s legal to be a passenger in a SDC but not operate one.[/li][li]Grandma who can no longer drive at night due to poor eyesight.[/li][li]A person with some disability or injury that prevents them from driving. One can stretch out across the back seat with a knee brace/leg cast but can’t physically sit in the front seat.[/li][li]Someone who is traveling a long distance & is sleeping during that overnight trip. They may or may not wake up when the car stops & wants them to take over.[/li][/ul]
All of these vehicles now create road obstructions when they stop & there’s no one in them able to drive.
For the second time in a week there’s a flood watch in this area. What I have received is for four states along the I-95 corridor; that’s not to say it doesn’t extend further north/south of what I received. We’re talking millions of people covered under this watch. You can’t have SDCs not work for millions of people on a weekly basis because of conditions that may be dangerous.
I don’t know why you think that this is such a difficulty. Why would placing it in the bumper (as I did not say, I said grill) make it not have sufficient time for stopping? This is the most trivial of matter, the location of a sensor. If it really is that big an issue, and a bumper one can’t see fast enough, and a higher one can’t see far enough, whatever that means, then they can put in two. Problem solved.
With the number of complaints about “The Pothole” that has materialized at the end of one of the local interstate exchanges, and the damage that people have complained has happened to their car due to it, it seems as though the eyeball MK.1 is poor at detecting and avoiding them too.
Note your parentheses. Yes, theoretically competent. There are quite a few that are not. These are the first that I’d love to get off the road, the 20% that really are not good drivers, whether they just aren’t smart enough to understand and follow the rules of the road, or they are not attentive enough to focus on the hunk of metal they are directing at what really are some pretty ridiculous speeds, if you think about it, or they are drunk, or sick, or tired. There are some times when I am not, whether that be due to exhaustion or illness, or just being distracted by a pretty bird. So, the 99.9% of the time that you don’t need a competent driver in the car works out great, and the .1% of the time, there is a small chance that there may be an inconvenience, due to not having a competent driver on board. If it is problem with the car or sensors, it can pull over until another car comes. If it is a problem with the road, then it will not want to go down that road unless forced to in manual mode.
Absolutely worst case scenario would be that you have to surrender control of the car to a remote operator (a switch that I would make as a manual switch that cannot be turned on automatically) who would be very highly skilled and navigate you through.
I suppose even worse case scenario, someone actually comes out in a car with manual controls, and you get in, and they drive you.
If the road conditions are such that the car has given up trying to drive, then there are already road obstructions that need to be dealt with by other cars. If possible, the car should pull over to the side of the road and ask for instructions. If not possible, then you have the same exact scenario as if a competent driver had come upon an impassible stretch of road.
For the second time in a week there’s a flood watch in this area. What I have received is for four states along the I-95 corridor; that’s not to say it doesn’t extend further north/south of what I received. We’re talking millions of people covered under this watch. You can’t have SDCs not work for millions of people on a weekly basis because of conditions that may be dangerous.
[/QUOTE]
That sounds like a problem with the roads not being safe. Are you saying that the I-95 corridor is washed out and under water? If so, then no one should be driving. If not, then I’m not sure why an SD would not work.
The sensor needs to be angled down to see something (not there) below the level of the road surface, the pothole or washed away road. Just as you can see further standing than sitting, or on a ladder than standing, the higher that angled sensor is the sooner it will see the hole in the road. I think a bumper is too low for a sensor to see something & have time to react to it.
so now your SDC needs extra sensors (connectivity to remote human driver, video camera for them to see what’s going on, etc.). A lot of companies don’t even include a spare tire for cost/weight savings; you think they’re going to include expensive stuff like that?
You know how long you need to wait for roadside assistance in bad weather?
I suspect there will be many smaller SDC, many not much bigger than an enclosed motor cycle of some type. They probably won’t have much ground clearance as they are fully on road vehicles.
I wouldn’t take a Ferrari off road, even to a grass parking lot, nor would I want a Hummer for a long highway ride. The single passenger car might have to give up, but enipla’s 4WD with higher ground clearance could make it thru just fine.
I’ve seen the same accuracy issues reported with that. Also, urban canyons are known to cause issues with GPS signal reception between not a clear view of the satellites & the signal bouncing off buildings on the way to the ground.
By I-95 corridor, I mean the roads within x miles of I-95 in (at least) 4 states, not necessarily the interstate itself but all of the local roads. However, there is another local interstate that is known to somewhat regularly be partially/fully shutdown in severe precipitation due to flooding/mudslides.
With regard to Differential Global Positioning Systems (DGPS). Yes it’s getting much, much better. Even in mobile vehicles. 10cm accuracy on a moving vehicle is a bit of a reach, but not for long.
I mentioned earlier in this thread that the head of Googles autonomous cars said that they won’t ever be able to work in all conditions (I’ll look for the link, very slow here now). The head of Waymo has said the same thing. That doesn’t mean it’s a bad idea. Just a bad idea for some.
And if I may interject. I think that Spiderman is not saying that a SDC won’t see a flooded road, but will not be able to see how deep it is. Sure, there are idiots that drive into 3 feet of water, but many, many more just slow down, and drive through 1" deep water. In controlled conditions, perhaps today, we could use LIDAR data to have the SDC computer range to the surface of the water, calculate its elevation and compare it to the LIDAR for the surface of the road and calculate the depth of the water at 50mph. Data, and processing power is pretty cheap. But if your data is bad, I’ll count on the MK1 eyeball every time. The MK1 is how we find bad data.
I’m a GIS programmer. 30 years* now in GIS. GIS Geographic information system is the study of relationships of one object or area to another (in it’s simplest terms) it’s what I do.
[sub]* When I started in GIS, back in '89, it wasn’t called GIS. It was called AM/FM - Automated Mapping and Facilities Management. When I saw AM/FM awards on our walls, I was a bit confused. I thought that we must have a hand in broadcasting.[/sub]
I’m a bit late to this thread. But there’s 3.1 million truck drivers, at an average annual pay of 42k, plus another million drivers working for various taxi services and others. SourceI used, other sources have different estimates. Without wasting time to look up the exact pay, let’s just assume it’s 35k. Another factor is that self driving vehicles have better capital utilization - truck drivers sleep in their cabs generally, as this is cheaper than a motel. This means that a piece of capital equipment that costs roughly 80-150k, average of 120k, is sitting idle while the driver sleeps.
Anyways, just replacing professional drivers - not all the other services that you could potentially offer, or factoring in the capital savings or insurance savings - this is a pot of money that has 165 billion dollars in it. That is, the autonomous driving industry will have to share some of that revenue with the trucking firms and taxi firms who fire 4 million people, so they do it - but it’s at least 165 billion. (the 165 billion number is potential revenue, not profit - you have to get an ROI on the tens of billions spent to develop the tech, plus actually manufacture each autonomous vehicle loaded with expensive sensors and computers, plus a maintain a substantial software team to maintain the software)
Americans in total drove 3.2 trillion miles a year. Get 10% of those miles carried by autonomous vehicles, and earn 10 cents on each of those miles in profit over operating costs, and that’s 32 billion in potential profits. 320 billion if you can collect it all and make personally driven vehicles outright illegal.
Anyways, this is the reason the vast efforts are going to develop an autonomous car. This, and other expected benefits, once the technology is developed for this purpose it is expected that the advances in machine autonomy will be applicable to many other fields. It doesn’t matter if the average joe wants to drive or not. There’s plenty of revenue replacing all the professional transport drivers to make a profit.
And the way these vehicles are being offered for personal transit is just like uber, but it will be cheaper. Did your car break? Are you drunk? Can you not afford a car right now? It get repossessed or you are banned from driving or have a medical condition or are too young to drive and mommy doesn’t want to take you to school? Not sure you want to pay $400 a month for the next 5 years for a new car just yet?
Just pull out your phone, ios or android doesn’t matter, go to the app store, remember the first few letters of the name of the app (one reason why uber is only 4 letters), hit download, give them a payment method, press summon.
Oh, you hate autonomous cars? Well, it only will cost you a couple bucks to try it this one time…
This kind of extreme convenience is expected to lead to rapid adoption, especially if the price per mile is not just cheaper than Uber but works out to be cheaper than car insurance + fuel + car payments for most American’s routes. (if you have to drive large amounts of miles per month it won’t be for you)
Spiderman, this isn’t how autonomous car positioning primarily works for actual vehicles being worked on. GPS just gets you in the ballpark. Once the car knows roughly what “grid square” of the earth it’s in, it looks up an onboard map. That onboard map essentially is a gigantic list of unique oddities that previous cars have seen around the vehicle. These can be unique visual features, but every other sensor the car uses can contribute. These are called landmarks.
As long as the car can pick up a sufficient number of landmarks, even if some of the landmarks are incorrect, it can work out the most probable position to within inches. This works all the time, everywhere, for places a previous autonomous car has been to contribute to the database. Note that the database is a lot smaller than it sounds - it doesn’t store the actual images, it stores various extracted features about the image that take up far less space. Here’s a documenton the algorithm.
With all this said, the car ideally only uses the mapping data to know where it should be located in the lane even if it can’t see markers, and what the traffic laws are even if it can’t see any signs, and where exactly it is, even if it’s in the back of some parking lot that isn’t on the map. Even if the mapping data is totally wrong, the car should still drive in a safe manner and prioritize direct dangers over the map or what it thinks the law is. (if it’s going down what it thinks is a divided highway and there’s an oncoming car, it should dodge it, not think it’s doing the right thing because it’s in what it thinks is a safe lane)
Good points SamuelA. Especially in regard to commercial vehicles.
As I said, I’m in GIS. I understand a bit how these databases work. But it still makes me chuckle that my Wifes car claims the speed limit on our barely maintained gravel road is 50mph . Not sure who’s data Subaru uses, but, it’s a bit off.
So sure, the system will recognize bad data that that can’t be right compared to it’s current local input and shut down. The cars themselves will need to be the input into these systems while they are using them. And will also have to weed out bad data which shouldn’t be too hard.
Currently, my Wifes 2019 Subaru Ascent is having a hard time keeping it’s radio and voice controls working. It can go dark on you at any time. You can boot that system. But that doesn’t always work. Next day, it works fine. Hardly a life or death situation. But it gives me pause with taking the control of the car away from the good 'ol MK 1 eyeball (I like that term, I think k9 owns it).
So yes, the technology kind of exists for this to work in very limited applications. But, like all new things, it’s gonna take a while to get it figured out and make it dependable. And then, there will be the thousands of things that we haven’t yet even contemplated that will cause problems. I am bringing personal anecdotes to this thread explaining where I don’t think SDC’s will work.
If you’ve read the other thread (which this thread is sorta starting to look like), you might recall that I’m skeptical of the combined timeframes/levels of autonomy that have been announced in the last three or four years. That’s a conclusion from having spent a fair amount of my work time on the topic, and from the fact that we’re spending less time looking at it than we used to. Something is definitely coming, and it will absolutely be disruptive, but it’s just going to take way longer (ioho) than was originally thought…I agree with where Gartner has put Level 4 and Level 5 autonomy on their hype curve.
Some thoughts on trucking, in response to SamuelA. I haven’t seen anyone suggest that AV will replace even a majority of trucking labor costs, with the main reasons being that eliminating long-haul trucking jobs would still necessitate hiring up for short-haul truckers, as well as building new depots where they don’t already have easy highway access. There is also the issue that we are a long, long way from putting L4 or L5 trucks (or even cars) on the road where whether can be inclement, which is still the majority of national highway. In any case, the estimates I recall are replacing 10%-30% of labor costs, but I don’t remember the details of what drove the haircut.
The nearer-term opportunity is for platooning, which is having a lead driver in a vehicle, with other vehicles tethered to that lead vehicle, and an off-highway depot sort of acting like a railroad depot, with local-haul truckers departing from there. I’ve seen a couple of test-runs of this, but I don’t know why that couldn’t be up an running in a year or two, once regulators are aligned [okay, just getting regulators aligned may take more than a year or two :)].
I was thinking that you put it around where the hood ornament would be (if you remember when cars used to have hood ornaments[whatever happened to that, anyway, just fuel economy, or too many of them getting stolen?]). On the front of the car, but at the top. There is no need for putting it all the way back on the hood near the windshield, and no need to put it down as low as the bumper.
It should already have all of those sensors for driving anyway. It should also already be connected to the fleet’s HQ and service. I just like the idea of putting in a manual switch to prevent the unlikely event of people being electronically hijacked. Also, probably “STOP” button or switch, for if the passenger is not liking at all what that car is doing.
About as long as I need to wait for roadside assistance in bad weather now?
Individual pods make much more sense than the 4.5 seat 2-ton sedan we usually think of as the minimum to be a car. Much better economy, and take up less space in traffic.
And that is another option. You could have models that are made for the purpose of being more viable with poor road conditions. Bigger tires, more robust transmissions, maybe a bag of kitty litter in the back.
There was a thread a while back about a car that went through a couple inches of water, and their air intake sucked up water, and it destroyed their engine.
The constellation has not been completed. They only have a few of the satellites they are planning on using. (And at least two of those are not really in the right orbit for positioning either.) Not sure what reports there would be available to determine the accuracy of the completed constellation.
And that is where differential GPS is useful. You can put transponders on top of all the buildings, if you wanted, and get millimeter accuracy.
I guess my point in that is that people are stupid, and they will drive through water, even if they don’t know how deep it is. Most of the time, it works out okay. Sometimes, it damages their car. And sometimes, they go floating off down the river.
A SDC will do what every driver should do in those conditions, and not proceed through. And, in the relatively near future, with better sensors, an SDC can actually determine with greater accuracy than a human can just how deep it is.
Nope, just renting without paying. As far as I can tell, the term goes back to at least the 50’s.
You’re mistaking the plans of companies who have nothing like the resources of the big players (Waymo, Cruise) who are working specifically on automated trucks with the capabilities of the big systems.
Waymo and Cruise, once they reach the expected point in development* where they have close to flawless level 5 autonomy, will directly making self driving trucks that are capable of all phases of the drive. No platooning or highway depots required. This is already being demonstrated, Waymo vehicles are navigating from parking spots at Walmart to pickup and dropoff points as we speak. I don’t expect the difficulties of big rig movement to pose any problems, at least compared to those already solved, backing up these vehicles or shifting gears are difficult tasks for humans but trivially easy for a computer. (bigger problems might be issues like needing cameras on the trailer or the different sensor positions making the data collected by big rigs incompatible with the other autonomous vehicles)
(*whether this is 2021 as stated in their business plans or 2030, it’s reasonable to think that this point is reachable)
This is where you and I take the Y in the road. If a car tells me I can’t do it, when I know I can, it will be the last car of that type I buy. And I might set it on fire for the warmth as that is as much good it is to me.
. Not sure who’s data Subaru uses, but, it’s a bit off.