Telematics, Driverless Cars and the Future of Transportation

[3278]

We talked over on Animalball about telematics and automated driving, but I thought this TED talk made some interesting points about the future of automated driving, namely something I've been worried about for a while now: that cars will get so good at the task of driving themselves that humans will no longer be allowed to drive them. As he says, though, one vehicle-caused death, of the right nature at the right time, could set back progress in this area dramatically.

Personally, while I'm all for using telematics and expert systems to automate the driving experience, I don't want to see the day come when I'm no longer allowed to drive, and I do believe such a day will come at some point. I also think that it will be 5 to 10 years before any system will be able to autonomously drive with the ability of an average human in all conditions; I'm thinking largely of night-time driving and driving in inclement conditions such as ice. Cars will be taught to deal with these conditions, and will be given senses far superior to our own with which to negotiate them, but it will be a while before our expert systems can handle, say, a nice Michigan winter as well as I [an average Michigan driver] do.

[paladin2019]

I would counter that this will never happen in the US precisely because of liability. A human will likely always be required onboard and responsible for whatever the machine does. That's my gut feel anyway.

I'm a little mre excited about the prospect of rail coming back. As long as the trains are car and gun ready, they'll probably be my preferred means of long distance travel.

[UncleJoseph]

At some point, I believe some systems will be fully automated. Trains, for example, are already starting to implement this concept. I'd especially like to see it in the area of cargo transport. If you look at other modes of transportation, the liability usually lies more with the human factor of the equation: The pilot or driver. Most aviation accidents are caused by pilot error, as are most automobile and boating accidents. Most of those accidents are a result of poor judgement, however, and programming "judgement" into a machine is difficult. So while humans excel at making decisions based on a constantly changing set of variables, we often make the wrong decision. Machines also don't have ego (yet, anyway), and many of the poor decisions made by humans are due to ego.

I think there's a strong argument on both sides of the equation. I'm not sure where I sit.

[3278]

I would counter that this will never happen in the US precisely because of liability. A human will likely always be required onboard and responsible for whatever the machine does. That's my gut feel anyway.

I'm not going to say "never," because that's a long while, but I do agree that even fully automated cars will, for quite a while yet, require an onboard driver. In other words, I can't send my car to get its own oil changed, but as long as I'm sitting in it, I don't actually have to do anything at all. I'm just there with my foot on the brake, just in case.

Of course, I'm reading a book and talking on the phone, but I'm legally liable.

Of course, we're actually already to that point: fully automated cars are driving on our roads with human drivers there only to push panicbuttons. That's Google's system, for instance, which drives on the public roads all the time. [Their goal, I think, is a fleet of drones which drive the nation, doing the equivalent of StreetView, but with many, many more sensors.]

But there will be a point, I believe, not too far in the future, when we do allow the machines on the road without us. It may be delayed by legal concerns in a variety of countries, but I think we've all noticed that, for the last 50 years or so, law has had a very difficult time keeping up with technology. I think to some degree we don't realize how badly we're doing at keeping up, because it seems normal to us that, for instance, there's a ton of stuff that's illegal to do on the internet, that hundreds of millions of people do, with no fear of prosecution. It's not impossible that driverless cars will sneak in the backdoor like this, beat the law to the punch; it's also not impossible that law will continue to get in the way of adoption, many years after that adoption would be reasonable.

These cars will be better than us very soon now. I'm not sure we're legally ready to admit that.

[Salvation122]

Airplanes are already better than pilots, but we still have pilots. And flying is simple compared to driving in traffic.

[UncleJoseph]

And flying is simple compared to driving in traffic.

It's the landings and take-offs that are tricky....

[3278]

And while our domestic passenger planes do have pilots, we do have planes without them. There are in fact a wide variety of unmanned aircraft which we legally allow to fly, people just don't ride in them. That day may well come, too, but considering the ratio of pilot to occupant, it's probably a luxury we'll afford long after it's technologically obsolete.

[Salvation122]

And flying is simple compared to driving in traffic.

It's the landings and take-offs that are tricky....

Most pilots still do takeoffs and landings manually, but modern autopilot systems are capable of doing the whole thing with minimal input on most approaches.

And while our domestic passenger planes do have pilots, we do have planes without them.

Cargo planes as well, which comprise a fairly large majority of air traffic. An MD-11 has a crew of (if I remember correctly) four regardless of whether there are live people or boxes in the back.

Edit: Two crew on an MD-11, three on a DC-10, three on older 747s. I guess I counted the trainer's chairs in the sims as crew.

[paladin2019]

I think it's mainly a matter of, "If something can potentially kill me in an accident, I want someone in control who can die in the same accident." Sort of adds an incentive to not have the accident in the first place.

[Bonefish]

Between this and other robotics, I predict the end of my specific career path in the next half century. Or, if not the end, definitely a big shake up. The first competitor to adopt driverless package cars is going to pull huge profits in, while UPS will struggle to keep paying it's expensive drivers( a UPS driver makes about 30/hour after 3 years, and generally pull long days). I'm not really sure how I feel about that, at all. On one hand, it'll suck, but I should be close to retirement by the time it really starts hurting. On the other, Hurray for progress!

[paladin2019]

Bonefish brings up the next obvious question. What do we do when another x jobs are eliminated by robotics replacing humans? At what point do we say screw it and just put the entirety of the working class on the dole? How long after until management is obsolete? How long until VIKI takes over completely?

[3278]

Most pilots still do takeoffs and landings manually, but modern autopilot systems are capable of doing the whole thing with minimal input on most approaches.

Man, I watched a documentary a few years ago that showed a modern passenger airliner taking off, flying, and landing, and the pilots explaining the process, and it was crazy: literally all they did on this Transatlantic flight was operate the throttle on takeoff and landing. I had no idea how automated these things had become: the pilots - and they're still absolutely essential in this role - are there as parachutes in case something goes wrong, and they represent the best current model for transportation automation: to each their own. The computers are better than the humans for the day-to-day consistent operations, while humans are better in nonstandard situations, so we let each do what they're best at.

For a long while, humans were better than computers at day-to-day operations, as well, and so it didn't make sense to have the computers do that, but now...well, the computers take less coke, for one thing. But when the day comes that the computers in airliners are smarter than the humans in all situations, then the days of the pilot are numbered: I'm sure they'll keep them around for a while for nostalgia and passenger reassurance, but eventually that salary is going to be the difference between their success and their competitor's.

And while our domestic passenger planes do have pilots, we do have planes without them.

Cargo planes as well, which comprise a fairly large majority of air traffic.

Absolutely. I don't mean to suggest that only domestic passenger planes have pilots: the vast, vast, vast majority of everything in the air is piloted, whether constantly or at-need, and that's not likely to change in the next handful of years. It may outlast, bizarrely, that situation on the ground.

So my point isn't that only passenger planes are piloted, but that we do finally have aircraft which are completely pilot-free, which is a major leap toward the future I'm envisioning.

[3278]

But it's not all news of online rights: rapid advancements in automated systems and telematics have been stacking up in my reading list, too, some of which are interesting enough to share:

Something I've talked a bit about before is inter-car communications; not between drivers, mind you, but between the cars themselves. Let's say I'm following you and you slam on the brakes to avoid a deer: today, some cars would use radar to detect your car slowing, and hit the brakes, but that reaction is going to come with some delay [and radar doesn't always work, particularly in adverse weather conditions]. What if, instead, your car told my car you were slowing, and mine automatically braked to match? That type of communication is now going to be tested in six US cities. V2V communications systems like this are helpful now, but will be absolutely crucial to the future of automated vehicles: it simply takes too long for your car to tell some central server that it's slowing, and then the central server to figure out who's behind you and tell them to brake. Instead, our cars will communicate like a swarm or school, reacting to conditions and to each other, without relying on a central server. Very cool.

Putting systems like this to the test, a Dutch competition was recently held to determine the best platooning V2V system, which gangs cars together as swarms to let cars pack closer together safely, and avoid traffic-jam-causing slowdowns and pileups. Almost as interesting as the competition itself are the /. comments on it, which overwhelmingly support technology over human fallibility. Of course, that's the crowd at /., which isn't exactly representative of the population as a whole, but which is often aimed in the direction society is heading, technologically. It's interesting how quickly we're becoming, as Steve Wozniac puts it, pets of the machines.

Perhaps slightly more disturbing for those of you who drive modern vehicles is a proposal from the NHTSA [PDF] which would require "event data recorders" [EDRs] in all passenger vehicles. Now, the intent of these is no different from an airplane's black box, but they've already been used - some cars already have them: yours may well be one of them, whether you're aware of it or not - to prosecute crimes recorded on the black box. Combine this with centralized telematics and V2V communications and all the other things we've been discussing, and you're getting perilously close to that terrifying future I've been envisioning in which these devices are simply used in real time to make sure you're doing what you legally ought to be behind the wheel. Setting that aside, this legislation would be useful in standardizing the type of EDR and the information gathered, which would be of extreme utility in making use of the data. But, as Wired notes, we have to strike a balance between utility and privacy; where we think that balance point should lie has changed dramatically over the years, and I look forward to seeing where it will head in the future.

[3278]

The funny thing about self-driven vehicles is that no one's exactly sure whether or not they're illegal. It hasn't been an issue. But it is, now. So Nevada - for reasons which aren't clear to me* - is blazing the way for the framework of laws in this field, by requiring their DOT to work up standards for them. It's not exactly a bold first step, but it's not exactly a shuffle, either.

Some manufacturers, though, don't want to jump wholly into automation. They'd like to investigate stopgaps. I hate stopgaps. Last week, the two most egregious examples were this Porsche and this Volkswagen: the Volkswagen can handle steering, brakes, and acceleration; the Porsche only brakes and acceleration; but neither is reactive or intelligent in any way, so the driver has to remain attentive and ready to intervene at all times. Yeah, right. I'm driving along in my car at 80MPH on the roadway and I don't have to have my hands on the wheel? I'm reading a fucking book, or talking on the phone, or turned around talking to my kids in the back. I'm not ready to take over control.**

These partial solutions are, I think, going to be dangerous. They give a false sense of security, tempting you to think your attention isn't necessary. But I thought that about adaptive cruise control, and it seems to have done okay. But I would prefer to wait another five years for full autonomy, rather than fill the gap with dumb systems that delay adoption of the superior technology, and which are themselves technological dead-ends. It's good for capitalism, but it's terrible for the smooth advancement of technology.

One thing I think is very interesting about the Porsche article, though, is the talk of using advanced sensors and expert systems to optimize efficiency. This won't by limited to dumb systems, and in fact will reach its full potential in fully automated systems: being able to gather data about conditions from thousands of vehicles, being able to very accurately map elevation using millions of sensor passes, being able to use this information to carefully tune the acceleration profiles and fuel-air mixtures of the vehicles: these kinds of developments will allow us to wring the greatest amount of efficiency from our energy, letting us do more with less. That's great potential, right there.

I'm not sure how I feel about Ford's V2V system, which basically lets cars talk to each other, and if it detects a couple will collide, beeps a lot. On its own, I'm not sure this technology is useful enough to justify the costs of massive adoption. But as a part of the future telematics infrastructure? It's a good tool to put in the box.

*Apparently, Google asked nicely.

**Not "I'm" as in "me, personally." My car can't even go 80MPH without the risk of massive suspension oscillation, so you can bet I personally don't do anything but drive when I'm driving.

[Cash]

Most pilots still do takeoffs and landings manually, but modern autopilot systems are capable of doing the whole thing with minimal input on most approaches.

Man, I watched a documentary a few years ago that showed a modern passenger airliner taking off, flying, and landing, and the pilots explaining the process, and it was crazy: literally all they did on this Transatlantic flight was operate the throttle on takeoff and landing.

Not any more...the plane can do it all...provided that the airport and the plane are equipped with the appropriate system, so older airports and older planes won't have this capability.

[3278]

*sigh* Seriously? I understand the Cylons haven't attacked yet, but why is it that everything wireless seems to just not bother with anything but the most basic security? I suppose probably from a dearth of user knowledge or give-a-fuck: "online privacy" wasn't exactly a watchword until everyone freaked out about Facebook.

Worse? The same technique works on SCADA systems, meaning you can hack stuff like, say, the water supply for your city. C'mon, vendors. This is as ridiculous as e-voting booths with exposed USB ports.

[paladin2019]

*sigh* Seriously? I understand the Cylons haven't attacked yet, but why is it that everything wireless seems to just not bother with anything but the most basic security? I suppose probably from a dearth of user knowledge or give-a-fuck: "online privacy" wasn't exactly a watchword until everyone freaked out about Facebook.

Worse? The same technique works on SCADA systems, meaning you can hack stuff like, say, the water supply for your city. C'mon, vendors. This is as ridiculous as e-voting booths with exposed USB ports.

Welcome to SR4 and hacking the bad guys' smartguns

[3278]

That was always my argument for why wireless technology wouldn't have been widely deployed in Shadowrun in the first place, but not everyone agreed.

Similar SCADA exploits can be used on systems using the same PLC Stuxnet targeted...and guess what PLC some prisons use? *sigh*

[3278]

Some of the concerns I've had about autopilots revolve around the atrophy of driver skill when it's no longer regularly used. The principle of many of the intermediary systems - radar cruise control, automated lane maintenance, etc - is that the machine will do what the machine does well [long constant activity, quick response to known situations] while the human can then do what it does well [good decisions in unknown situations]. Except that a recent report suggests that, in the case of airline pilots, the atrophy of skill has surpassed the ability of the pilot to make good decisions in unknown situations.

And I think this is exactly what'll happen as we add more partially automated system on the road: the car will drive itself 99 percent of the time, and increasingly humans will do stupid shit the 1 percent of the time they're asked to do something, because driving a lot is how you get good at driving. In theory, this will lead to a pressure to get rid of those systems, but if the timing is right, the pressure will be not to increase manual control, but to eliminate it entirely, and ban drivers because, statistics will then show, they're really awful at driving.

[Bonefish]

If driving a lot is how you get good at driving, how come so many fucking drives suck?

[3278]

Well, there's more involved than just that, unfortunately.

[DrunkenMaster]

As far as skills atrophy goes, I wonder what level of maintenance is needed after you've gained proficiency. I know in certain types of workouts, after you've put in the hard work, you can maintain your gains with a 3 or 4 day a week program. I also see the same thing with other knowledge/skill based tasks such as basic mathematics or marksmanship.

I think you can overcome the atrophy of a skill set with a maintenance type program for telemetics or piloting airplanes. I'm certain there is research on this, but I'm lazy.

Maybe 10% of the time human controlled and that 10% is divided throughout the week or month. Just enough to keep your mind in the game and engaged, and you can focus specifically on the tasks you want to achieve instead of zoning out and going through the motions like typically happens when doing a repetitive non-engaging task.

[3278]

Well, that's something else: supposedly - and I have no reason to doubt this - airline pilots are required to constantly re-test for their positions, with difficult and constantly varied tests that have to be studied for. Aren't these guys doing simulator time, doing emergency maneuvers, doing the things they'd need to do to not freak out then they stall?

Anyway, yeah, it seems to me that, for pilots at least, some percentage of manual time should be required. Not sure how this could work for drivers, but that seems like a simple enough solution to the pilot issue.

[Pdyx]

If driving a lot is how you get good at driving, how come so many fucking drives suck?

I read an article in Scientific American (or some such thing) about talent (basically saying that almost all activities aren't inherent skills, but require dedicated practice) and it used this example (driving). I wish to shit I could remember the details, but it indicated something about increasing in skill requiring increases in challenges as well as just practice time. We aren't all as good of drivers as Mario Andretti not only because he logs more hours driving but because he is actively trying to get better by having increased challenges, coaches, people watching him and analyzing him, tape watching, etc.

It used that example because even if he may drive more than us, but we drive a lot (or some people do). Just because you engage in an activity isn't going to make you better at it. You have to practice with increased challenges. I think it used Chess as an example. If I can beat you in chess, and you're the only persno I play, I probably won't get any better than I need to be to beat you. But if I then start playing other people better than me I might increase my skill.

There are undoubtedly ways to become a better driver that don't even require additional practice other than normal commuting but you'd have to actively engage in that.

[3278]

Joseph and I have been saying this for years, although maybe we were just trying to justify reckless driving as "practice." Still, I credit limit-pushing with whatever minor skill I might have behind the wheel.

[Pdyx]

I came across this article today, with much relevance to this thread:

http://spectrum.ieee.org/automaton/robo ... -car-works

My apologies if some of this has already been covered, but it's a new article about the Google driver-less cars.

[Salvation122]

Well, that's something else: supposedly - and I have no reason to doubt this - airline pilots are required to constantly re-test for their positions, with difficult and constantly varied tests that have to be studied for. Aren't these guys doing simulator time, doing emergency maneuvers, doing the things they'd need to do to not freak out then they stall?

They are. They have very good simulators. The FAA considers Class D sim time to be equivalent to actual flight time for training purposes.

Anyway, yeah, it seems to me that, for pilots at least, some percentage of manual time should be required.

Pilots usually do take-off and approach/landing manually. On most planes they have to.

[3278]

As mentioned previously, any scheme designed to guide vehicles or ascertain road tax using GPS is going to have serious problems.