January 9, 2008

It's Not A Flying Car But...

...it may be the next best thing: automated cars. It seems like something everyone would want, right? You wouldn't have to be stuck doing nothing while sitting in traffic, it could help alleviate pollution and accidents by driving more efficiently, and it could save you time! The idea has been around forever in science fiction, but whether you realize it or not, it's been a reality since 1989 with the ALVINN project at Carnegie Mellon University. That little guy, which learned how to drive by "observing" other drivers that pilot it, has been putting around the CMU campus by itself for years.

Larry Burns, vice president of R&D at General Motors, seems very optimistic in the AP article linked above, claiming the driverless cars could be in production and on the roads by 2018. While it's a nice thought, I'm pretty sure that this isn't going to happen. The system described in the article works by networking cars together to help them avoid each other, but the problem with this is that you're not going to be able to roll out this system all at once. Not everyone is going to be able to immediately trade in their current car for a driverless one, and there are going to be a lot of people who won't even want to. So, not only is this system going to have to be able to network with the other cars on the road that contain this system, it's also going to have to be able to drive without being in communication with most other vehicles on the road. That problem alone is going to add several man years to the current description of the project.

That, in effect, is the biggest problem this project faces. Since the driverless mode on the vehicle is optional, the biggest obstacle is going to be the human motorist. Computers, by definition, are logical, and humans, by nature, are not. The vehicles that are being controlled by computers are going to be driving in the most efficient way possible, and this is going to drive the human motorists who prefer to control their vehicles themselves absolutely insane. Allow me to give you a few examples:

  • A computer will always obey a posted speed limit. This fact alone will make human motorists' blood pressure go through the roof (especially if they're from the DC metro area), as these cars won't be going "fast enough." Granted, if enough of these cars hit the streets, the speed limits could be raised, but the human drivers will still want to go that extra five to ten MPH faster.

  • If a computer senses a light turning red, it's going to just let off the gas and coast up to the red light. It's not going to burn unnecessary fuel speeding up to the light and then hitting the brakes hard. This will infuriate those people who think that getting to the red light faster actually decreases the amount of time they will be spending in their vehicle.

  • A computer is also not going to slam on the gas when a light turns green, either. It is going to slowly accelerate in the most fuel efficient manner. I can hear the horns already.

These are just a few things off the top of my head that I know will annoy human drivers, and possibly even human passengers, regardless of how efficient the process might be in terms of fuel efficiency or obeying the law. I'm sure if you think about it, you'll be able to come up with some good ones yourself. Of course, there are some motorists who live to infuriate impatient drivers1, and they will absolutely love all of the aspects mentioned above.

In addition to the aspects above, the automated driver will also lack the ability to view and interpret the body language of a human driver. While driving down the road, a human can glance at another driver, process an insane amount of visual input in less than one second, and determine that another driver is impatient/aggressive/otherwise dangerous and take action to avoid this person.2 I don't see artificial intelligence reaching the ability process this amount of visual information efficiently in the next century let alone in the next ten years.

Of course, all of these issues I've mentioned boil down to the fact that the automated driver is an option instead of a requirement. I think the only way that this system is going to work any time soon is that they are going to have to slowly phase in cars that have support for this system. Once everyone has them, they will be able to activate it and phase out human-operated vehicles. Once activated, you're not going to be able to enter a manual mode without a damn good reason, and how the car is going to determine whether a reason is acceptable is a very good question and yet another roadblock to this project. The only other way this is going to be feasible in the near future is if huge advances in artificial intelligence are made. Unfortunately, I just don't see that happening in my lifetime.

1Umm, so I've heard. I wouldn't know any of these people personally, mind you.
2Or, if you're one of those people, see exactly how much you can piss them off.


Courtney said...

The other main problem, I believe, is that automation isn't so good at processing in three dimensions (specifically up-down). So it would be very good at staying between the white lines, but would be likely to drive off a cliff (or get stuck on a curb, for a less morbid example) if given the opportunity.

There's also the added cost of updating the road systems, under the current proposed ideas (sensors instead of or in addition to road lines). Estimates are upwards of 10M miles of roads in the US, one sensor every 15-20 feet...adds up quick.

Matt Silverthorn said...

Well, solutions exist for the first problem you mention, but they would definitely require work. For example, the iRobot line of home robots, including the Roomba, have sensors to detect when they are approaching ledges. The problem with cars is that you'd possibly be doing this at 60mph. So, yes, it's something that would have to be fixed.

One of the possible solutions to the first and your second problem is GPS navigation. I'm not sure how accurate it is right now, but f they can get it to the point to where it's accurate to within a few inches or so, that could help solve this problem as well. Now, I'm certain that commerical GPS like Garmin is not this accurate, but I bet the military is using technology that has this level of precision.

Matt said...

I think we'll see self-driving cars in our lifetime, but I'm not sure I like the possibility of having self-driving cars along with human-driven cars. Mostly because there is a lot more to driving beyond "don't run into the other person." Would the car know that on most of 495, you need to go at least 5-10 over, even in the rightmost lane? Or when allowing traffic to merge, does it know to balance between letting people in and maintaining position? If it maximized fuel efficiency, then it would attempt to coast as much as possible in this situation and you'd get taken advantage of by many drivers.

Courtney said...

Right, the Roomba can detect a ledge...when it's approximately 1-2 from it. But in a passenger car at 65mph you're going to need at least 200 feet of braking distance to stop safely (assuming we remove the human response time as a factor in stopping) on DRY pavement.

Which also begs another question, how will an automated car take weather factors into account and reduce speed/increase following and braking distances accordingly? How would it compensate for a fishtail or spin if the car hydroplanes or hits a patch of ice?