It seems to me that a self-driving car would be next to useless, if the best it can do is emulate the driving style of each user. The main purpose for self-driving is to improve the efficiency of the roadways, and to do so with more safety than you would expect out of mere human drivers.
For example, there are plenty of drivers out there who become catatonic at red lights. They waste valuable seconds waking up to the fact the the traffic light turned green. This sort of waste creates traffic congestion (because fewer cars get through the intersection during that cycle), and requires wider and more roads to be built. Ideally, with self-driving cars, this sort of inefficiency would be eliminated, to make existing roads more capable of handling increasing demands. Ditto with freeway driving. Human reaction times require huge spacing between cars, at typical freeway speeds. A real waste of road capacity.
My thinking is, instead, that self-driving cars will be tailgating each other, at speeds way beyond what a human could manage safely. And jamming through intersections with way much tighter tolerances than mere humans can master.
I think that internal sensors in the car will be used to check the health of all critical systems, and that any sign of impending critical problem will automatically shuttle the car off to an emergency lane. Less urgent matters will create a warning for the driver, to take care of whatever problem soon. Of course, the latter is already true today, so no big vision there. There will also be road sensors, of course, to warn of obstacles or other road conditions, to adjust the flow of traffic.
>> "Why shouldn't my car speak with an Italian accent?
The problem with many technologies is that public infrastructure will limit the progress or the speed of innovation. If you live in Pittsburgh with all the potholes, you will know that govt will be unable to provide the ecosystem to enable some of these ideas to flourish. And this is American-wide.
These are also my concerns. Here in Germany many towns and cities haven't got enough money in order to remove the street damages of last winter. How should they then ever finance the needed car-to-x infrastructure?
The Google car can tell whether the light is red or green. Emergency vehicles cause the light to change. It seems like a self-driving vehicle handle, say, a cop car running a red light as well or better than a person can. The Google cars have over 500,000 miles on public roads with no accidents. They're getting better every day.
Good point indeed Olaf. I think this technology is coming at the wrong time in history when nations are broke to do any major project financing. Pot holes and bad bridges, I wish Google and the likes good luck if they expect govt to help make this mainstream.
You aren't alone in being skeptical about the premise of self-driving car.
The success of driverless cars, in my opinion, actually depends on a critical decisioneach carmaker makes -- in terms of how they design a car that can interfere with a driver's freedom. Hence this discussion of "personalizaion."
As I wrote before, it has less to do with engineerig problems, but much to do with if auto engineering can understand human psychology.
This gets into an interesting area of engineering/technology implementation versus the marketplace and consumer psychology. On the one hand, as Bert22306 suggests, personalizing self-driving cars to emulate their owners would seem to defeat their purpose - or at least completely undercut most of the potential advantages. On the other hand, how many consumers are likely to buy into the concept of essentially purchasing a taxi or bus for their personal use, where the driving itself is completely automated and out of their control?
Perhaps these self-driving cars will come equipped with robotic or virtual chauffeurs? Then the perception wouldn't be a case of buying a "self-driving car" but instead a car equipped with a "personal driver."
Will the self driving cars take all the fun of self-satisfaction of control? Today also people who can afford have chauffeurs, which actually is synonymus to the self-driving cars. As far as technology is concerned, i think we have it in pieces but the biggest missing piece is the infrastructure.
The first goal should be to make a car that you can drive as you always have, but that will not let you run into another car, person, or object, and that will not allow you to lose control and kill yourself or your passengers.
Once the car does that, you can drive it any way the hell you please within those constraints. All the way from doing everything yourself, (just within the constraints of not hitting others or losing control), all the way up to allowing the car to do all the driving while you sleep in the back seat. I'm not sure why we couldn't have it all -- except for the part where cars kill you or others around you.
The first goal should be to make a car that you can drive as you always have.
Yes, I think that's the key. By extention, though, if you are letting your self-driving car drive (but bear in mind that you might need to jump in when the situation gets complex), you don't want the car to override your instinct, either.
With the self-driving car, whether you like it or not, I am guesing that eventually there will a time that your car interferes wtih your own driving. I think this "handover" iussue is a much bigger one than any of us realize today.
I think the "handover" issue is something we already have -- one way -- with Cruise Control. There, a tap of the brake or pressing a button on the steering wheel releases control to the driver. But we don't have the reverse, where the car suddenly leaves cruise control on its own, assuming the driver is ready to take over -- that would be dangerous.
So I can't quite imagine a car that is not only controlling the speed, but steering as well. If that released on its own without the driver initiating the handover, chaos would ensue. But, according to this article, cars can't handle all sorts of driving situations on their own yet. Until they can, I think this technology must remain in the testing phase.
The handover mechanisms need to be carefully worked out. In the simpler cases, i.e. where the cars are driving alongside manually driven cars, it seems easy enough. Just like cruise control. The driver can take over just by grabbing the wheel. Since humans are driving out there, the margins have to be very wide. No problem.
Seems to me that in the more interesting applications, like on roadways dedicated to self driving cars, you wouldn't want to allow a human to take over. These automatic control systems invariably bring the technoloy to where humans are incapable of safely operating the machinery. So in this case, if the human needs to get out urgently, you need to command this to the car. And the car would then need to coordinate that operation with the other traffic.
Not sure why so many seem to assume that human control is the ultimate safety feature. There are countless control systems out there where any type of human in the loop would be devastating. Maybe, at best, humans might intervene through special software control. It's not so inconceivable that driving will eventually reach that same place.
Bert: I think the reason folks are worried about giving up control to the machines is because there's no assurance they will be better drivers than the humans -- and you bring up an interesting example with the thought about a human who needs to urgently exit the car.
Let's say there's a fire inside the passenger compartment -- we've had some horrid fatal incidents of that recently in the SF Bay Area. A human would draw on all mental capability to stop and get out of traffic very, very fast. But giving a command to a computer to stop the car might not bring the same sort of panic stop. I'm guessing the car would follow normal protocol of signally surrounding vehicles and reducing speed gradually. I suppose you could build a panic button into the dash, but that raises all sorts of other questions....
Sure. For one thing, a self driving car will be infused with sensors, because it is imperative that the car know its own health at all times. I'm talking about these cars using the roadways efficiently, not just a half-a**ed approach to self driving. So this car will sense the fire starting faster than a human would, and either put it out, or take action to stop the car immediately.
But also, why assume that a human driver can stop a car faster than this algorithm-driven car? Humans do all sorts of irrational thing when they panic. We already know that automatic controls do many many jobs much faster and more accurately than humans can, so why wouldn't that also apply to emergency stops?
Technology seems to evolve this way all the time. During a transitional phase, that comfortable well-known alternative is kept as backup (like sails were retained on steam ships, and hand cranks were retained on cars with electric starters). Pretty soon, though, those old standbys have to go. You can't even use a hand crank when the compression ratio is higher than 5:1 or so. Imagine having to live with that restriction. So the old standby becomes liability. That's what will eventually happen with human drivers.
Bert: Well said, my friend. I'm sure that someday it may be possible to have self-driving cars, although right now most cities can't seem to fill the potholes much less build new roads. I wonder how long this transition will take...?
I recall hearing serious talk of self-driving cars for the first time at the NY Worlds' Fair in 1964 -- almost half a century ago -- at the GM exhibit there. They gave us the impression we'd all have them by the turn of the century. I'm guessing I may never see them in common operation.
How long do you think it will be before we see millions of self-driving cars in operation? And how long will it take to switch from them being mixed with human-piloted cars to the day they own the road themselves?
I think we are overly simplifying the dynamics associated with our central nervous system. The human brain with its association of sensors spanning our entire body is one mechanical system that is very hard to model and control. There are just too many degrees of freedom. If we confine the control problem to a restricted, well defined or well-known domain with a small finite number of unknowns, then it becomes increasingly realistic to design a control system to work within the boundaries.
This is what most control systems are about today. They have several output variables which they attempt to keep within a control limit. This is possible with combat aircraft, submarines and many other advanced control solutions we have developed. However, in all of these cases, you still need that human factor to account for the unknown.
Self-driving cars will be operating in a uncontrolled environment with a lot of unknowns. This alone makes it a very difficult proposition. Like someone mentioned, we are barely doing a good job in maintaining our relatively primitive transportation infrastructure. How will we fare with such a highly intelligent and advanced traffic infrastructure? We are automating certain tasks in cars today but at the end of the day, we still need that human factor to cater for the unknown.
A similar argument can be applied to robots. There has been significant progress made in this field but there is still a long way to go before a robot can undertake a sample portion of the tasks we are capable of. It must also be said that some emotional or irrational human decisions could sometimes end up being the appropriate ones based on the context, something which is impossible to automate.
I would certainly rather work on my laptop during my commute (1 hour each way) than worry about driving. On the other hand, I do like listening to my audio books. But the biggest snarls in traffic are often from some stupid fender bender (or worse) and if self driving cars can do nothing other than keep space between the front of the car (i.e. stop and/or speed up), this could be a good thing. Would definitely take some getting used to...
Along those lines, when I drove by a google "self driving" car, I really wanted to go in front of it and slam on my brakes to see if it would stop in time. Am I the only one who wants some personal data acquisition?
The Google cars are driving around every day continually making the universe of unknowns smaller. If they double their current miles, they'll have demonstrated that they're safer than the average driver. Without changing the infrastructure.
"The car must emulate the behavior of the individual driver"
This is actually to be avoided in the self driven cars. The majority of accidents happen because of this individual behavior of the drivers especially when the action of a driver mismatches with the reaction of the other driver ( coming opposite, tailing, or in the adjoining lane)
If the self driven cars decide themselves how to drive in a given situation ( traffic jam, sudden braking by the car in front, road block, slippery road, snow etc) then actions and reactions of the two cars adjacent or opposite to each other will match and the accidents could be avoided.
Don't make a self-driving car that follows personal behavior...was also what my instinct said initially.
But the chat with Freescale's Mr. Santo opened my eyes. Think about it. Self-driving cars, in many ways, will interfere with the freedom of your own driving.
Of course, if self-driving cars are designed strictly for those who loathe driving, it's a different story.
But many drivers today have their own instincts, sense of freedom, which they enjoy and cherish while driving.
If self-driving cars turn out to be a big turn-off for those drivers, they won't go anywhere...
Sorry, Junko, but if people cherish their freedom and their own driving experience, then they don't want to travel on those roads that will be set up for self driving cars. Otherwaise, you defeat the prupose of the self driving car.
Obviously, this would take a very long time to develop. At first, only a few roads would be set up with the necessary infrastructure. So people should have options. But quite honestly, I see this in a larger context, where more and more of the driving has been taken away from the driver, in the past 100 years.
Used to be the driver decided on everything, from changing gears to setting the choke, fuel/air mixture, and the spark advance, to covering up the radiator on cold winter mornings. Very, very inefficient. As things became automated, in each instance, it was SPECIFICALLY to take away that individual human input, and to have the job "done right."
Self driving is no different. Take away the quirky and unpredictable human behavior, in order to make traffic flow smoothly and efficiently. Just like you took away the driver's ability to set the fuel/air ratio, in order to let the combustion process be as efficient and clean as possible.
But I am not certain if there will be a separate road designated for self-driving cars. Where did you get that info? I would like to check it out. My understanding is that self-driving cars do not necessarily require so-calle infrastructure, although car-to-x infrastructure could presumably provide more intelligence to all drivers.
"Moreover, it was connected wirelessly to the Internet, giving it access to a vast cloud-based set of data that could be matched to what the local sensors were seeing."
It makes sense. It's very doubtful that local in-car sensors can be counted upon to know that there is road work at that intersection 5 miles down the orad, and therefore you should use a detour. Or that even that the lane next to you, which appears empty, is only empty because a mile or so down the road it is blocked off for some reason. things that people might find out from a radio report or the morning news, or other such source.
Self driving cars can't constantly be blind-sided by things that their on-board radar/lidar/what have you, would have a tough time knowing or predicting.
I suppose they can mingle with manually driven cars, although that takes away a whole lot of potential efficiency. Now you have to leave in big margins to accommodate erratic human behavior.
I think there are a number of interesting issues here. I think we can all agree (or most of us anyway) that ADAS has already improved vehicle safety and has the potential to do a lot more as it proliferates. But it's one thing to have this technology as an enhancement, and quite enough to take the driver out of the picture altogether.
First off, we are many many years away from this being a reality. And even when/if it becomes a reality, the driver is still going to need to be involved. One danger I see (among many) is if there comes an emergeny and a person needs to take control of the car in a split second, that driver had better be paying attention. You can see the potential for the "driver's" attention to wander if he or she is not actually driving the car most of the time. I think that is a huge issue.
There are any number of examples of technology having taken over a previously manual function, specifically because human control was not fast enough. I gave some simple examples in cars, but this sort of gradual change is pervasive in everything. As the years go by, and the technolgy continues to evolve and improve, it becomes almost unthinkable that the job was ever allowed to be done by a human previously.
In the car example, catalytic converters would not survive a day (or at least ,not very long), if the choke and the fuel/air mixture were still trusted to humans. In fact, they wouldn't survive very long even if these jobs were still done by mechanical carburators. That's the main reason why all cars had to go to electronic fuel injection. And we hardly give that a second thought anymore.
Stick shifts are gradually going the same way. Too unpredicable and inefficient. Many modern stick shifts coax the driver into the shifting to the "proper" ratio. And too, for efficiency, gearboxes are now sporting 7 or even 8 gear ratios. Practically impossible for a human to select the right gear.
Driving is probably going to evolve the same way, over the next many decades. And I wouldn't be a bit surprised if people in the next century will marvel at how humans were EVER trusted to do their own driving. Sort of like wow, you must have really been naive to trust the other guy to drive next to you.
Bert: I think you're on the right track. I was a stick-shift driver until once, when buying a new car, I realized the automatic trannies actually got better gas mileage. When technology becomes better than what we can accomplish on our own, then we hit a tipping point that leads to broad acceptance.
IMHO, we're at least a generation from that tipping point with self-driving cars. But perhaps one day....
Several people in this thread have made the point that drivers vary in their responses to situations, but that also applies to driving itself. The car manufacturers like to think that drivers enjoy the act of driving so much that they would not want to give it up, but I would say that varies significantly from driver to driver and from time to time. I'm a big fan of automatic climate control, because I can just set it to a temperature and forget about it. My wife, on the other hand, constantly fiddles with the temperature and fan settings because the automatic system just never gets it right in her opinion.
Any long-time commuter is also aware of the difference in traffic patterns between rush hour and random times on the weekend. Rush hour is the closest that we currently have to automated driving. People tend to fall into very predictable patterns during this time, and even bad ideas tend to be standardized. I will tend to pull over and let BMW drivers pass, since in my experience they are more likely to tailgate, and people will tend to be more aggressive about getting around me in my Honda Insight, since they know that I tend to accelerate more slowly. Weekend traffic, on the other hand, gets much more random.
My expectation is that many commuters would gladly hand over the controls during their commute, but some might be less interested in doing so over the weekend. Eventually behavior will change, but that usually changes much slower than the technology.
LarryM99, I just laughed out loud about your wife constantly fiddling with the temperature and fan setting of air conditioner. I am with her!
But you do have a point when you said:
The car manufacturers like to think that drivers enjoy the act of driving so much that they would not want to give it up, but I would say that varies significantly from driver to driver and from time to time.
BMW may be overthinking of their brand -- billed as "sheer pleasure of driving."
"...a self-driving car will eventually learn how you drive -- everything from steering to acceleration and other driving habits." seems like a way to resolve the issue of liability when accident happens.
Leaving liability and responsibility aside, I can see the future of self driving car. Even though it may still be a concern in traffic, I am pretty sure long distance driver will enjoy it very much. If nothing else, cargo transportation company will be extremely happy to adopt the technology if government allows self-driving car to go on the road w/o a "driver".
Having said so, I still believe liability and responsibility to an accident are the biggest stumbiing block to the wide acceptance of self driving car.
Fascinating (and scary) stuff! I was particularly struck by this line:
"There will be times," Santo explained, "when the situation on the road gets too tricky for a driverless car to handle."
So, let me get this straight. I'm in my auto-driven car, tooling down southbound I-5 at 3 am at 95 mph, looking forward to my weekend of debauchery in Tijuana, when my car gets freaked out by something it senses in the road ahead -- maybe a cat, or a Greyhound bus. So it stirs me from my inattentive haze and says: "Here! You Drive!" (And then I probably panic, crash and burn.)
Good to hear, Bert. But from the story, it sounds like they're still trying to figure that part out. And I don't want one of those cars on the highway with me until they do. (I've driven for 4 decades without an accident...knock on wood.)
Tom, I am with you. I got scared when I heard that during the interivew, too.
But I think what he is talking about is a much more complex scenario. Self-driving cars shouldn't freak out at the site of a greyhound bus or a mere cat. But in a much more complex urban setting, you can easily imagine four or five things going on at an intersection. (picture a major intersection in Manhattan)
Rather than waiting for your car to make a decision, you might want to take over the control of your car.
Years ago now, I had the interesting experience of driving a submarine control system emulator (designed for training crews). This was a new device at the time. A real eye opener.
Unless you've driven a sub manually, you may not realize just how tight the allowable depth tolerance is. If you're moving fast, just a little too close to the surface, and the sub is likely to broach (break the surface) unexpectedly. Conversely, just a little too much down plane, and at high speed you'd surpised how fast you get to crush depth. It's not like these things can be turned around on a dime. It takes lots of training and skill to do this safely, and one would only change depth at low speeds.
Now turn on the automatic controls. No problem at all. You can go at flank speed, command max depth, and it'll go there with no apparent strain. Go fast at periscope depth, also no problem with broaching. It all seems so easy, yet it's close to impossible to do with humans at the controls.
Ditto with controls for those experimental high performance foward-swept wing jets. A human can't control these airplanes, except with control system software assistance.
I'm a long time science fiction fan, and the notion is an old one in the genre. The usual expression is that you get in the car, tell it where you want to go, and the car takes over from there. You sit back and watch a movie or something. In general, you aren't driving, The car is, You're simply a passenger.
Achieving that state will not only require sophistication in the car, but a network infrastructure the car connects to. It will be the network's job to decide how to get to the location I want to get to, picking the best route, and managing the traffic in the process. The process would be analogous to TCP-IP packet routing, where the exact path a packet will take to a destination is not specified, and may vary from packet to packet depending on conditions in the network
While manual control fallback would likely be an option, all concerned would want to make it as unlikely as possible that it would ever need to be used.
Self driving cars will come in the not so distant future, no doubt. We have already seen prototypes, which find a parking lot by themselves. We have adaptive cruise control, which is based on radar and we have lane departure warning and blind spot detection based on cameras and image processing already since a few years. The step towards (semi-) autonomous driving in stop and go traffic up to a certain speed is not a big one. Gradual improvements will finally bring the autonomous car, which drives at higher speeds and passes slower traffic. No need for changes in the infrastructure, all done by optics, radar and processing power. After all, who would fund such an infrastructure? Maybe Asia, but the western countries are on a decline, all of them. They don't even have the money to maintain the current infrastructure (e.g. bridges). Who needs (semi) autonomous cars? Well, we all may need them when we get older and become handicapped in one way or another. Remember that, also in the western world, many people live in rural areas without much public transportation. My father suffered from a stroke earlier this year and he had to give up driving. With his children living far away, it is difficult for him to get to the doctor, who is a mere 5 miles away. When I'm in his age, I will enter the car and tell it to bring me to the doctor. It doesn't have to be fast and driving needs not be fun, but it will eventually get me there. And regarding the fun side of driving, there are enough situations, in which driving is no fun. A six hours ride leaves enough opportunities to lay back a while for a break or to drive manually with enthusiasm. Just like on a cross-country flight in a small aircraft, where even the passionate pilot leaves flying to the autopilot.
Some of the scenarios you described in your post perfectly capture the real-word's needs for autonomous driving. I feel for your father. And yes, we all need someday a car that can get us to where we need to be.
Meanwhile, if not a complete self driving, we do understand the need for semi-autonomous driving. For safety, it's indeed a great help.
And yet, what is not so clear to me is to get a handle on when and where my judgement is required while driving vs when and where I can totally trust my car to do the job.
Blog Doing Math in FPGAs Tom Burke 24 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...