Hardware

8 truths and myths of driverless cars

As we look toward a future with automated cars, it's time to separate the science from the fiction.

googlecar.jpg

For decades, humans have dreamed of driverless vehicles. From the Jetsons to Minority Report, we've gotten a certain idea of how those cars should function and how the world could be if they existed.

Driverless cars, however, aren't science fiction.

"When people get into our cars, we often find that they're unnecessarily anxious about the experience, and then we find that they relax too quickly," said Chris Gerdes, program director for the Center for Automotive Research at Stanford (CARS, for short). Neither extreme is helpful.

That said, we took a look at some of the common misconceptions about driverless cars, along with truths that are good to keep in mind.

Myths:

Driverless cars eliminate human error

The happy idea about automated cars is that they would totally eliminate human error-- no accidents from texting, drunkenness, sleepiness, or even momentary inattention -- but in reality, according to Gerdes, automated vehicles shift human error from the driving to the programming and design. "That can actually be a really good shift, but it's not one to take for granted," he said. "The idea that you could take a step back and program the car in the comfort of your office, in a lower stress environment to handle all these stressful situations is potentially a huge improvement in safety."

Where the risk comes in, he said, is in failures of imagination. In order for a car to be able to handle a situation, the programmers have to have envisioned it, or something similar enough to it. "If you think about all the weird things you've ever seen while driving a car, and multiple by about 200 or so, that might be what you would get in 100,000,000 vehicle miles traveled," he said.

For example, Gerdes was driving home from a Stanford v. Oregon game last year when a person jumped in front of his car. He slammed on his brakes and the person turned, and ran up the hatchback of the car in front of him, on to the roof, jumped off, and right back into traffic. "Imagine trying to come up with a perception system that understands where pedestrians might be. Have you programmed it to look on the roof of cars?"

He said it's a challenge that's both inspiring and humbling -- the sense that they could take life and death decisions, and turn them into programming challenges.

Humans are bad at driving

You'd be forgiven for disagreeing-- some days people drive like they're trying to hit each other -- but Gerdes still identified two exceptions to the idea that humans are bad at driving. Take race car drivers, for example. "Their ability to get to the absolute limits of what a car can do is, at this point, unparalleled." Experience, intuition, and the solid ability to use friction help to deal with an emergency quickly. "When you look at designing the these emergency maneuvers, at the moment, the best humans are still better than the automated system," he said.

The other advantage that humans have is how they understand their environment. It amounts to context and perception. "That's something that computers are still struggling to equal. The 360 degree scanner might be more efficient than a human, but the picture is still not as clear because it lacks the depth of understanding.

Automated cars can drive anywhere

Greg Fitch, research scientist at the Center for Automated Vehicle Systems at the Virginia Tech Transportation Institute said many people believe that Google, for example, has built a car that can drive anywhere on its own. "The car itself can only work in a very limited context right now. It has to be very good weather, it can't handle parking garages because it can't get a GPS signal. The reason why those cars can drive themselves is because they know where they are in the world," Fitch said.

Passengers can be passive

An automated car isn't an automated cab, Fitch said, even if that's the eventual trajectory. He said that despite systems that keep a car in its lane, for example, manufacturers are expecting that car owners will be paying attention with a hand on the wheel and a foot near the brake.

However, Fitch said, the problem that's starting to crop up is people not only being passive, but find ways to trick the car's systems, the ones created to make the cars safe. In one instance, Fitch described a YouTube video of a driver who had taped a soda can to the steering wheel in order to fool the car into perceiving that he was holding the wheel. Another guy managed to climb into the backseat of the car.

"Believe it or not, that car could fail at any time," Fitch said.

Truths:

Automated cars can be more energy efficient

"If you look at how much energy we use move the person today versus actually moving the car, when you have humans surrounded by a couple of thousands pounds of steel, you're probably spending on the order of 85% of the energy to move the steel and not to move the person," Gerdes said. The reason for most of that steel is crash protection, but if you consider vehicles moving at lower speeds that could avoid collisions, the need for so much extra material diminishes and energy is more efficiently used to move people.

There's an ongoing ethical debate

There are numerous questions still to be answered: How conservative should the car be to avoid accidents? Should cars be able to brake the law and speed in order to keep passengers safe? Humans, after all, make those decisions every day. Gerdes also brought up Asimov's laws, and if there's room to add to them.

"Asimov had protecting human life as a higher priority than obeying human orders, so all of our cars that we drive right now have this big red button, which the human driver can take over at any time," he said.

However, if you're going to implement something similar to Asimov's laws, it could complicate things. For instance, if the car felt an accident was imminent, it might not let the driver take over because that would be responding to a human command at a higher priority than saving a human life.

There's no centralized governance over driverless cars yet

So, who decides these debates? That's still in the works. Government agencies and states are working with the issues, there's also a standards committee from the Society of Automotive Engineers, which is looking at creating a set of voluntary standards for vehicle design, Gerdes said.

There's a debate over whether cars should be connected

Fitch falls on the pro side of the argument. If cars can communicate with each other, they can share emergency messages or basic info like location, speed, and heading. Human drivers can be inefficient in terms of how they use the road. If a car knew what was around the corner, like another car, roads could be used better because cars could drive with less distance between them. The opposite view says that human drivers aren't connected and can drive with just their eyes, so automated cars should, in theory, be able to operate with just camera vision.

Also see:

About

Erin Carson is a Staff Writer for TechRepublic. She covers the impact of social media in business and the ways technology is transforming the future of work.

24 comments
David J Laurie
David J Laurie

Will take over realy quickly as good technology does. People seemingly resist change, but young people arent so jaded as us oldsters and accept change. People will look back in wonder and awe that we manuslly controlled out autos. Also, once we can begin true space travel to Kepler habitable planets, which number in the billions in our Galaxy, the Milky Way, and in the hundreds of billions s in our SuperCluster, Laniakea, there will be very few abortions because we will need and want and value each human life for settlements.

Muyinda Sam
Muyinda Sam

In a country like mine (Uganda) where potholes (of all sizes and shapes) are kinda normal; automated cars would require very high levels of A.I to learn and adapt to the way automobile drivers manouver their way on such poor quality roads. Anyways, I think it would be fun getting to teach the cars how to adapt to our driving practices in Uganda

Alan Portman
Alan Portman

I have been a pizza delivery driver for almost 30 years. It will take more than a decade before the mapping software gets close to being usable in any but the most mundane situations. I have seen Google, MapQuest, and Bing routing software in the last 5 years. Bing has a 30% error rate. Google was the best but still routes through blocked off streets and will go onto a highway, up 3 miles and then back, rather than cross the highway. Come see me when the car can actually find where it is going. I don't even want to think about emergency situations. No, lets. Autopilots are orders of magnitude better than car routing. The pilot still lands the plane. The pilot still routes around weather.

bradhansen
bradhansen

"Automated cars can be more energy efficient" -- that's a no-brainer. At least twice a week some incompetent driver passes me as I'm coasting to a red light. Often they pass me in the left or right turn only lane before cutting me off and slamming on their brakes. Then they wonder why their vehicle doesn't get the gas mileage that was on the sticker.


Infrastructure-to-car and car-to-car communication would greatly enhance efficiency. E.g. for infrastructure-to-car, cars would automatically coast, then brake when their projected arrival at a light would occur when it was not green. Car-to-car would eliminate stop and go traffic on interstates (or any road with few traffic lights) as cars would "know" the optimum speed to drive. Yes, that optimum might be only 5 MPH, but at least it would be essentially constant, as opposed to starting and stopping which greatly lower fuel mileage.

DenebQuarlo
DenebQuarlo

Of course the cars MUST be connected. Connected by proximity, road, destination, etc...

I get tired of these articles that suggest only mistakes can be made and competence is only located off shore!

DenebQuarlo
DenebQuarlo

Who the bleep is Fitch? Am I wrong or did that name just appear in the article without reference?

dapaine
dapaine

REALITY:  "The reason why those cars can drive themselves is because they know where they are in the world," Fitch said.


QUESTION:  What will happen to the market for driverless cars the first time one of these vehicles rolls up in front of an embassy or consulate or shopping mall or sports stadium or other high-value target and the 500 lbs. of ANFO in the trunk is detonated?

tony
tony

A few points


The EU has targeted zero road deaths by 2050. Probably won't be achieved, but interesting to see how close we can get.


I used to design the hardware and write the software for automotive systems. You try and think of as many scenarios as you can, but you will never think of them all. So no system will be foolproof. What you have to do is to balance the risks, probably asymmetrically. By this, I mean that the gain to the population at large (the common, or greater good) is at least an order of magnitude bigger than the risk to an individual of a failure. Taking a safety feature, it would have to improve matter in say 90 accidents whereas things may be worse in 10.


If we think back to the compulsory wearing of seatbelts. There were lots of stories about people being thrown clear and miraculously surviving what would otherwise have been a fatal crash. But statistically there were probably 100 times more people saved by the safety belt.


I have had a Mercedes B-class with the radar cruise control for the last couple of years (it also has a dashboard warning of an impending collision). There is a bend near where I live with a road sign that will trigger the impending collision. This is a false positive. Using the radar display to detect what is in the road ahead of me, I can confirm what the manual says - it doesn't detect cycles, very small motorcycles, but does detect motorcycles with panniers. So it works well, but is not perfect, so as I a driver I have to use common sense and treat it as an aid to driving, and not a foolproof feature.


Likewise, you would think that the automatic gearbox (which is more economical than the manual) would give the best fuel economy left to itself, and especially on the highway with the radar cruise control. In fact, I can better it; the radar keeps a constant distance, but only to the vehicle in front. I, on the other hand, can see that the cars in front of the car in front are starting to slow down and ease off the gas ahead of time. So although the system does a really great job, and actually good at anticipating on lane changes and when faster cars cut in, it still doesn't have the wider capability of anticipation that I have. On the other hand, if I get distracted and fail to spot a hazard that it does, it will save me from myself.


No system will ever be 100% safe and perfect. We should think of them less as autonomous cars and more as greater safety enhanced cars. Autonomous cars have the same problem that good drivers have - they still can't cope always with the other idiots on the road, at least not until the cars take over 100% and none of us are allowed to drive or override the controls.


If you consider that the life of a car is (say) 20 years, then to be able to have only "safe" cars on the road in 2050, this means that we have to have those cars in place by 2030 (not so far away) and be prepared in 2050 to ban cars from the roads that don't meet those standards. Can you see that happening?

da philster
da philster

Interesting technology, but if in the quest of implementing this technology we end up "dumbing down" the driver, have we really progressed? I think that perhaps we should back up a bit and concentrate more on better driver training and testing as opposed to using chips to protect the inept that don't belong behind the wheel in the first place.

Randy Myers
Randy Myers

Probably will not work during a storm (sand, rain or snow and meteor showers.)

Keith Higgins
Keith Higgins

An unfortunate inevitability, as our freedom vanishes and government control increases. ..

michael
michael

One can just imagine the carnage at Calais when the Shuttle Trains and the car ferries disgorge the vehicles carried across the English Channel onto the French Roads, some will drive on the left, some will drive on the right, and many as now, will just drive in the middle of the road. If people get much lazier they will lose the use of their brains (assuming they already have one)!

kjohnson
kjohnson

In fifty years' time the argument will be about whether it is safe to allow a human being to drive a car on a public road. (It isn't: 3000 deaths a year is more fatalities than a medium size war.) Car parks will be equipped with navigation aids for cars, just as they are currently being equipped with sockets to recharge the batteries of cars with electric propulsion.  A good analogy is with tarmacadam. In 1910 nearly all roads were just made of puddled earth, and everyone complained about the choking clouds of dust that cars raised as they puttered along at fifteen miles an hour. Now, almost all roads have a tarmacadam or concrete surface.


For inter city travel, I expect that cars will drive themselves onto a high speed transporter, like the Motorail service that British Railways used to offer being privatisation ruined it. Cheaper and faster than using the motorway, and the train has comfortable seats too. At meal times, of course, the car will place and pay for a meal at a burger bar or a gourmet restaurant half an hour down the road, and then it will drive you to the counter so that you can open the window and pick your order up. On an overnight journey the car will doubtless close the shutters so you can doze off in the bed that converts into the back seats during the day.

Darryl Duke
Darryl Duke

"foot near the break."


Give me a brake...

Jesse Reinosa Segovia
Jesse Reinosa Segovia

Yes please! I would be so much productive if I did not have to sit in LA traffic for 1hr in my commute

DenebQuarlo
DenebQuarlo

It's travel would be tracked back to the source. The car shouldn't be able to travel down a public road without identifying itself. Even if it spoofed another car it could be tracked.

tony
tony

@michael  I believe that in the 1950s, there was an Italian car with the steering wheel in the middle and a passenger either side of the driver. There is a prototype car around that moves the steering wheel from one side to the other - possible to do with steer by wire. Especially for us Brits who regularly drive both sides of the English channel. Having had both LHD and RHD card both sides of the channel I drove all four possible combinations in a few weeks sometimes. I found it simple to ask myself two questions when I get in a car (1) am I the driver and (2) if the driver, do I sit next to the side of the road or the middle of the road. Then everything you do is a mirror image and it halves the things you have to think about. When my daughter was learning to write, for just a few weeks, if she picked up her pen in the other hand, she would write everything backwards. Thus, at one point in learning to write, her brain was commanding her hand to move away and back, and in and out, rather than left and right. Having observed the point that the brain starts with mirrored commands, it seemed logical to apply to my driving problem. And at least for me, it has worked for 30+ years since I started to drive all combinations of LHD & RHD.

TRgscratch
TRgscratch

@Darryl Duke   If " manufacturers are expecting that car owners will be paying attention with a hand on the wheel and a foot near the break. ", what is the point of building a "driverless car" ?

NickNielsen
NickNielsen moderator

@tony The latest models of the Unimog have driver controls that can be converted from LHD to RHD in the field.  It's neat, but I doubt the Unimog will be one of the first self-driving vehicles.

Editor's Picks

White Papers, Webcasts, and Downloads

Don't Miss Our Latest Updates