Would You Need a License To Drive a Self-Driving Car?

agent elevator writes Not as strange a question as it seems, writes Mark Harris at IEEE Spectrum : "Self-driving cars promise a future where you can watch television, sip cocktails, or snooze all the way home. But what happens when something goes wrong? Today's drivers have not been taught how to cope with runaway acceleration, unexpected braking, or a car that wants to steer into a wall." The California DMV is considering something that would be similar to requirements for robocar test-driver training." Hallie Siegel points out this article arguing that we need to be careful about how many rules we make for self-driving cars before they become common. Governments and lawmakers across the world are debating how to best regulate autonomous cars, both for testing, and for operation. Robocar expert Brad Templeton argues that that there is a danger that regulations might be drafted long before the shape of the first commercial deployments of the technology take place.

Comments

If "yes," then it's not self-driving

[xxxJonBoyxxx]

If "yes," then it's not self-driving.

Re:If "yes," then it's not self-driving

[Shakrai]

It should be down to the manufacturer to ensure safe, autonomous operation.

Thus guaranteeing that it never happens, at least in the litigious society known as the United States of America.

Aerospace is held to a far higher standard than automotive ever will be, with modern planes able to fly themselves from takeoff to landing, but we still expect qualified pilots to sit in the front seat and keep an eye on things. An autonomous automobile may well have more variables to contend with than an airliners autopilot. Children don't tend to dart out in front of airliners, the physics of air travel don't change drastically with weather conditions, and airplanes are built with more redundancy than automobiles.

Even if you can account for such things, how will your autonomous vehicle handle malfunctioning sensors? Aerospace has been working at this for decades and still hasn't figured it all out.

Re:If "yes," then it's not self-driving

[HiThereImBob]

Self-driving cars should be the legal equivalent to sitting in the back of a taxi. Even from an insurance/liability standpoint, owning one means you're responsible/liable for fuel & maintenance - and that's about it. It should be down to the manufacturer to ensure safe, autonomous operation. (Otherwise, things such as self-valet and timed pick-ups won't happen)

Let's be realistic. Self-driving cars are coming, but it is going to be a gradual transition. We've already seen the beginning of it with adaptive cruise control and self-parking. These features will continue to be refined while new ones are added, but we almost certainly face years (decades?) of gradual transition where our cars are some weird hodgepodge of self driving and user operated. The laws governing this won't be nearly as straightforward as you suggest.

Re:If "yes," then it's not self-driving

[kylemonger]

Forget about sensors for a moment: We don't deal with malfunctioning PEOPLE right now. Drunks, old people, and visual impaired people routinely climb behind the wheel everyday. We are already running over darting children, cyclists and pretty much anything else with the temerity to set foot, hoof or paw on the road. Old people ramming cars into crowds because they can't tell the brake from the accelerator are just the cost of doing business in a free society.

A self-driving system doesn't have to be perfect, it just has to be better than what we have now when we scale it up. Given that you can give a driving AI the equivalent of millions of miles road experience in all conditions, I doubt that AI's will drive worse than human beings for much longer.

The insurance companies will need to be convinced for sure, but they will be when self-driving systems demonstrate their superiority.

Re:If "yes," then it's not self-driving

[fuzzyfuzzyfungus]

Your suggestion about the pace of the transition is highly plausible; but that needn't imply much regulatory complexity: At present, a car with a licensed operator can have a variety of convenience features that involve a level of automation and (very) bounded autonomy without any variation in what type of license you need for what level of features. There may be some hassle on the vendor's end, in convincing the relevant feds that their intended new feature isn't an automated accident generator; but there's nothing on the driver side.

Given that operator handoff is most likely to happen either under relatively hairy conditions, or when some system failure has left the automated systems unable to cope, there isn't an obvious incentive to relax the(already not terribly demanding, at least in the US) requirements placed on licensed drivers until 'self-driving' actually does mean 'self-driving'. If it means 'sometimes self driving, except the hard parts', that may require less operator effort; but not obviously less operator knowledge(if anything, given that drivers usually get somewhat safer with experience, at least until they hit the point where each additional year stops making them less young and stupid and starts making them more old and inept, I'd be particularly worried about the likely performance of somebody whose vehicle is sophisticated enough to coddle him most of the time, then screams and hands him the wheel when the situation is already halfway lost.)

I have no doubt that the laws(or at least the liability litigation and insurance-related contracts, even if carried out under existing law) for damage and death caused by partially-automated vehicles will be an epic nightmare of horrendous proportions; but on the operator licensing end "If you might have to drive it, you need a driver's license; if you won't have to drive it, you don't." really covers a lot of territory. There might be some incremental adjustments, mostly to the format of the test(say, allowing use of a rear-view camera in addition to mirrors and over-the-shoulder during tests of parking); but not too much need to complicate things.

Re:If "yes," then it's not self-driving

[hawkinspeter]

I'd much prefer to share roads with autonomous vehicles, especially when I'm walking or cycling. They have the potential to have much, much better collision avoidance than your average driver. With a decent array of sensors (e.g. infrared, radar, lidar) they should be able to detect humans (maybe other animals as well) and reduce speed to allow corrective actions if necessary.

I'd anticipate that autonomous vehicles would be able to react a lot quicker and of course they wouldn't be distracted (driver distraction being the number one cause of accidents). In the case of an emergency, I wouldn't want the vehicle to be relying on the human to wake up, figure out what was going on and then take appropriate action.

Re:If "yes," then it's not self-driving

[wvmarle]

Even if you can account for such things, how will your autonomous vehicle handle malfunctioning sensors? Aerospace has been working at this for decades and still hasn't figured it all out.

Detecting a malfunction in a sensor is hard, really hard. You'll need more than one sensor, preferably different types, to realise there's an error, and then you have to decide which of the contradictory sensor results is the correct one. As naturally sensors will always return slightly different results, you'll have to account for that as well.

So let's say we solved this. Then you know there's a problem. For an autonomous car it's simple: it could decide to continue (minor problem), or stop (e.g. tyre blow-out or other major problem that makes it unable to continue, or simply "I don't know how to handle this situation, so I pull over to the side of the road and stop to have my human overlords sort it out"). In the second scenario an automated call to the repair service could be included, so the human(s) in the car can continue to sleep while it's being fixed and after that be sent on their way again.

An airplane doesn't have this fail safe stop option, and needs to have human overlords present at all times to take control if something happens the programmers didn't foresee.

Re:If "yes," then it's not self-driving

[wvmarle]

Given that operator handoff is most likely to happen either under relatively hairy conditions, or when some system failure has left the automated systems unable to cope,

Euhm... let me get this right... you expect cars to drive automatically, except when it gets difficult or something else unexpected happens it suddenly gives back control to the driver. That's what you mean, right?

Bad idea. Very bad idea. The driver is probably reading the paper, or is dozing off, or otherwise simply not paying attention to the road, as the car is doing the driving and he has nothing to do. He's not supposed to do anything about driving, as the car is in full automatic driving mode. Suddenly asking for attention, then expecting the driver to handle a difficult situation instantly, is asking for accidents. Many more than when the driver was in control already, and possibly sees the situation coming, so anyway has much more time to react.

To allow the driver to fully hand off control to the car, the car should be able to handle it all. The driver assist functions we have available on certain cars nowadays are a great start in working towards full control by the car: now the car will intervene in certain emergency situations, when that's all settled, we can think about giving off control of the rest of the ride as well. For fully automatic drive, the car should not rely on human intervention, ever.

Why....

[Anonymous Coward]

A self driving car should be able to be aware of it's surrounding, of the other cars around it, and if anything goes wrong, either set itself to the side of the road automatically or just Stop, while the others cars around that are also self driving, slow down and avoid the broken down car. Never requiring any action from the dumb*ss behind the wheel.

TL-DR

[wbr1]

As with any new technology, society and by proxy the government will struggle with its place and rules in said society. There will be oversteps, misses, and oversights.

There will be detractors, luddites, and evangelists, sociopaths and attention whores all vying for a moment in the sun.

Welcome to the human race. I'll go get my popcorn.

Do pilots still need licenses?

[mjwx]

Do pilots still need licenses in the age of autopilot? Well yes because machines aren't infallible.

For a long time, an autonomous car will not be driverless. People need to get over this notion that next year a car will drive itself and you'll sit in the back with a Martini and the paper. That probably wont happen in our lifetimes.

Initially, fully autonomous modes will only be permitted on certain roads (think limited access roads like highways, freeways and autobahns). This will last years as engineers are even more conservative than law makers. The next step is likely to be special lanes on A roads. It will be a long time before autonomous cars are good enough to operate on a B road or suburban street.

Ultimately, because the law requires someone to be responsible for the operation of the machine it means a qualified operator will need to be at the controls whilst in operation. Same with a lot of other automated systems (such as long distance trains).

Re:Do pilots still need licenses?

[FlyHelicopters]

Bullshit. Watch Air Crash investigations, there are enough autopilot caused incidents to show why pilots are needed.

You might think that, but you'd be wrong...

Pilot error is the single largest cause of all aircraft crashes...

The reality is, if you simply accepted that every time the computer messed up, everyone would die, you would have FEWER deaths than you would with the current system.

Because no one wants to hear that and can't emotionally accept any deaths as "known", we have what we have.

Source? 15 years of aviation experience, thousands of hours of flight time in everything from helicopters to corporate jets, Certified Flight Instructor in both airplanes and helicopters. I know better than most people on Slashdot about this subject. Just because you watched something on The History Channel doesn't make you an expert.

Re:Do pilots still need licenses?

[geekmux]

Sorry but chess is a well defined game with simple rules. Playing chess needs excellent memory and fast computation. Computers are very good at that. Driving a vehicle is very different and requires much higher and different intelligence.

Every time a computer gets good at a task once thought to be outside of the realm of AI, people simply rationalize how it wasn't that hard in the first place. Soon people will be saying how self-driving cars weren't that hard to create for whatever reason. Then something else will be "impossible" for a computer to do.

You know the funny part about your story here?

It's actually believable...in a universe devoid of lawyers.

We live in a world today where litigation is the main limiting factor with technology, not the technology itself.

After watching the birth of patent trolls as well as companies spending millions fighting for inventions like "rounded corners", I would have thought this would have been more obvious. It will be when vendors are sitting around for a decade waiting to deploy their inventions while we argue who should be sued when it fucks up.

Re:"Promise a future where we can sip cocktails"

[Anonymous Coward]

"Only fucking MORONS want this sort of thing.

When you're in a piece of heavy machinery, like a car, even if you're NOT driving it, you DON'T want to be impaired in case of an emergency."

Wha?? Hows this any different than taking a cab home?

Am I putting myself in horrible danger when i take the cab home after a night drinking? Afterall i dont want to be impaired in case of emergency. Friends dont let friends take a cab?? -rolls eyes-

Blah... you people worry too much.

Re:Insurance

[tmosley]

Fault is determined the same way it is today. You can bet your ass that self driving cars will have black boxes full of data on speed, location, orientation, etc. If one causes a wreck, that would probably be the end for that manufacturer from a publicity standpoint. As such, you can bet that any model that hits the road is going to be DAMN safe.

A horse for a horseless carriage

[jader3rd]

You would need a 'driver' for a driver-less car as much as you need a horse for a horse-less carriage.

Re:Really? Come on now, you should know better.

[Jeremi]

Nothing, and that is an absolutely nothing, has ever been made by man which has been perfect.

A self-driving car does not have to be perfect. It just has to be better than the alternative.

With motor vehicles already being the number one killer in the US annually, we want human intervention early and often.

Isn't the fact that motor vehicles are already the number one killer in the US annually actually an argument for automated cars?

As stated above, a half a century has not perfected "self driving" anything else.

Five centuries of work before that never perfected heavier-than-air flying machines either, until one year, presto, all the necessary preconditions were finally met and airplanes became a reality. There's nothing linear about progress.