Transterrestrial Musings

Comments

And if they use bio-computers, they may get drunk on ethanol.

Posted by Sam Dinkin at December 4, 2007 09:05 AM

I figure we'll be doing better on reliability when the cars can go down to the shop and get their oil changed, tires rotated, belts replaced, etc. on the proper schedule without any humans having to miss a TV show////////take time from their busy day.

Most likely the smart cars will have a "defensive driving" control which lets you set the level of paranoia you want it to have. So all the confident commuters will form a pack and blow past the more cautious ones like me.

Posted by Karl Gallagher at December 4, 2007 09:21 AM

If you have a swarm of cars operating in lockstep 10 feet from each other at 90mph on the interstate when one car fails, the other cars in the swarm will have to respond to the failure. If the cars are in constant communication with each other, this should be possible. If the engine dies without warning, the cars behind the failing car can know to slow down. If a car's breaks fail on a straightaway, the cars ahead of the failing car can know to slow and gently impact the failing car, slowing it down. If a car's breaks fail in a turn, things may be more tricky, but the swarm can still act to protect both the failing car and the rest of the swarm. If a tire blows and causes a car to swerve violently to the left or right, the rest of the swarm will have to instantly react accordingly. Bumpers would have to be redesigned, and ideally roads would have larger margins to accomodate evasive maneuvers.

Posted by Eric Goldstein at December 4, 2007 09:25 AM

There is actually quite a lot of improvement that can be made.

First, the key is networking. If the smart car just reacts to the car in front of it, it can improve slightly on human reaction times, which are at best 1/4 of a second and usually longer.

However, if the smart car can network with the 5 or 10 cars ahead, it can get advance information on the state of the road and react before a normal driver would get that information.

Consider a line of cars starting up as a traffic light turns green. The first car moves. About half a second to a second later, the driver of the second car notices that the first car is gone, and accelerates. By the time you are down to the 10th car, you have lost 5-10 seconds of the cycle in lost reaction time.

With networked smart cars, the 10th car could begin to move at the same time as the first car, maintaining a minimum 6ft spacing which could be programmed to grow as speed increased.

Traffic light cycle times could be reduced, or more vehicles accomodated per cycle.

However, that is not the real obstacle. From the Article...

"Smart cars will never be infallible, but they don’t have to be. They just have to be better than the drivers who now cause more than 90 percent of traffic accidents and kill a million of their fellow humans per year."

This may be right technically, but it is wrong legally.

Under the pernicious legal doctrine of strict liability, it is no defense that the machine you are selling is better than what existed and may save, on the aggregate, thousands of lives. It must act perfectly in every instance, and there will always be an expert witness to testify that you could have done it differently, or made a different design trade-off that would have kept old Mrs. Jones from getting into that particular accident that put her in a wheelchair.
And even though it is 95% her fault for not wearing her seatbelt, she is indigent and therefore you with your 5% liability have to pay all of the damages, including her care for the rest of her life, her pain and suffering, loss of income and punative damages.

We expect far more of our machines that we do of our citizens.

Posted by DocBrown at December 4, 2007 09:37 AM

I have to give the nod to Mr. Goldstein. Given how fast computers can crunch data, if the software is good (and I mean REALLY good), then the autobots (what else would we call them?) should be able to deal with any likely hazardous situation better than we do. Also, consider that mechanical systems rarely fail without some prior warning, which these cars would almost certainly be wired to detect prior to a failure point being reached. Tire technology is also advancing to the point where "blowout-proof" tires will likely be available and mandatory for this type of application (if not universally). If the software can be made reliable enough, I think this is a sure bet.

Posted by David A. Young at December 4, 2007 10:02 AM

As long as Microsoft doesn't make the operating system.

Actually, this is a Nanny Stater's dream. Look for it mandated by law circa 2050.

Posted by K at December 4, 2007 10:07 AM

The first problem with dealing with brakes failing is detecting the failure. Which can be kinda tricky before you attempt to use the brakes.

(For instance, an O-ring failure on the brake piston won't be easy to detect before pressure is applied, causing the fluid to squirt out...)

(And the difficulty with wiring everything to detect failures is not just expense, but complexity and the problem of detecting failure (including false positives) in the detection system. I'm not even sure how you'd try to detect an incipient hydraulic failure, though I expect there must be ways...

At some point, trying to be fool-proof, you end up building the space shuttle.)

Posted by Sigivald at December 4, 2007 10:26 AM

Under the pernicious legal doctrine of strict liability, it is no defense that the machine you are selling is better than what existed and may save, on the aggregate, thousands of lives.

I can see it working the other way around as well. As soon as the cars are demonstrably better in even a limited situation, people will start suing companies and/or individuals for failing to use or make available these vital "safety features".

Besides, the second the first baby boomers start losing their licenses due to old age, the demand for these cars will skyrocket. How else will they get to the Rolling Stones concert?

Also note that the same sort of lawsuits mentioned above could be used for airbags, and have. Still, airbags are now almost universal...despite the fact that they sometimes decapitate children, which is about as gruesome and jury swaying as it can get.

Posted by sjv at December 4, 2007 10:57 AM

As far as the wheels go, that's why we need to adopt a technology called the tweel. It makes blown tires a thing of the past.

Posted by taoist at December 4, 2007 11:17 AM

How would a tire blow when everyone uses run-flats and the sensor tells you that you have a problem long before it is a problem?

How old is your Beemer?

Posted by Offside at December 4, 2007 11:21 AM

How would a tire blow when everyone uses run-flats and the sensor tells you that you have a problem long before it is a problem?

How old is your Beemer?

Posted by Offside at December 4, 2007 11:21 AM

Let's see: we're assuming that the autobot will be able to accurately distinguish and correctly react to safety-affecting conditions such as water, mud, wet leaves, snow, strong crosswinds, glare ice, snow on glare ice, a wet road that is icy under the overpass, a sandy patch in the corner, water-filled pothole, a deer springing out of a ditch, tumbleweed in the road, etc., in temperatures from -40F to 120F, and in all weather.

Then for the “swarm” concept, we also assume that the networking between cars is secure and spoof-proof (wanna bet somebody would devise a jammer that would make all the other cars on the road automatically pull over?) and will be fully compatible between different cars that were built in different years in different countries, on different tires, with different handling characteristics and vehicle dynamics. And of course, as Rand notes, the highly complex software and actuators/servos will have to be ultra-reliable over the 20-30 year life of the car.

I think we’ll sooner see $10/lb to LEO, or even more likely, those Jetson’s flying cars.

Posted by Chuck at December 4, 2007 12:15 PM

"What happens if cars are barreling along at ninety miles an hour ten feet apart, and a tire blows?"

I think the same safety practices that NASA follows in these situations should apply. Safe the vehicle through destructive means. Wha? the loss of a few is for the good of the many :P

Really though, I'd think the number of track events would go up dramatically for people to get their kicks driving around without computer assistance. More circuits like the Nurburgring that you have to access some type of toll gate and be notified of total human control beyond this point. On the flip side there could even be amusement ride where people climb into a computer controlled high performance race bus and ride around a crazy 3 dimensional track.

"As far as the wheels go, that's why we need to adopt a technology called the tweel. It makes blown tires a thing of the past.
Posted by taoist at December 4, 2007 11:17 AM"

I believe one of the groups that is against these type of tires is law enforcement. One of their favored tactics in pursuit mitigation is to employ the use of spike strips to blow tires out. A trade off with this future technology of self driving cars would give law enforcement the power to instruct your car to automatically pull over as they see fit. Currently, police officers avoid areas of heavy construction or narrow shoulders since pulling someone over is just to dangerous for both parties. In the future they could put your car into a follow mode and lead you off into a safer area if need be. Which brings up the question of how speed limits would be handled. Keep in mind that speed limits are often set well below what engineers currently design roads to be safely traveled. Speed limits are also set under environmental and fuel efficiency concerns. So, just because the car drives itself doesn't mean 90mph speed limits would necessarily follow unless vehicles could provide equal levels of efficiency as at 60mph.

I think the reality lies somewhere in the middle though as far as how much control we will be willing to give up. The idea that the car would go park itself is certainly appealing to me. Be able to put the car into a dedicated high speed automated lane on the highway would be useful in certain circumstances. However, I love to drive though and I love cars and having that right to do so taken away completely would be just bleh.

Posted by Josh Reiter at December 4, 2007 06:27 PM

"I think the reality lies somewhere in the middle though as far as how much control we will be willing to give up."

The middle of where we are now and where the control-freaks would like us to be is always going to approach tyrrany as time advances.

But in any case, I don't think traffic could be controlled remotely. Air traffic control can barely handle air traffic as it is, and that's with no terrain considerations and a whole other dimension to work with.

If you want computer control of your cars, it would have to be done by the cars native computer and senses themselves. This would be for a number of reasons, not the least of which is that computer controlled cars will have to drive with non-computer controlled cars for a good long while (you know, a phase-in period), on non-networked roads, to non-vista-ready destinations (which, if they can't reach them, they won't be nearly as valuable to any consumer with a choice as those old-fashioned mechanical cars).

Another reason is that local-control algorithms (road rules) make sense for handling arbitrary traffic levels. They scale. Central control doesn't.

Furthermore, even if your computer had instant reaction time, it can't stop or avoid accidents that much further outside the reaction time of your typical human (1/5 second). That's because a car is a mechanical system with latencies involved in just about everything.

And a computer that is actually interacting with real noisy data from the outside world has latencies of it's own to deal with.

Posted by Aaron at December 4, 2007 07:31 PM

"I think the reality lies somewhere in the middle though as far as how much control we will be willing to give up."

The middle of where we are now and where the control-freaks would like us to be is always going to approach tyrrany as time advances.

Another reason is that local-control algorithms (road rules) make sense for handling arbitrary traffic levels. They scale. Central control doesn't.

And a computer that is actually interacting with real noisy data from the outside world has latencies of it's own to deal with.

Posted by Aaron at December 4, 2007 07:31 PM

Anyone else here ever watch recreational ice dancing? I don't mean the Olympics or World's, but where a whole bunch of men and women get together to skate all at once.

Unlike ballroom dance off the ice, ice dances skated for recreation in sessions with other people are rigidly choreographed. There is an intricate dance pattern of steps that happen on particular measures and beats of the music that fill up the entire ice surface, and you and your partner skate in lock-step to that pattern, and couples ahead and behind you are respectively a measure ahead or a measure behind in phase with respect to you.

When a dance is announced, couples line uo for launch like the flight line on an aircraft carrier in a WW-II movie. You even start to see them sway a bit in place as they get ready to get in time with the music, and from the front of the line, another couple initiates their introductory steps to merge into the dance pattern. One couple after another initiates their steps, each on successive measures of the music, until the entire dance pattern fills up with skaters.

You can focus your attention on specific couples to see how they skate, or you can gaze upon the entire rink full of skaters and observe how they are zipping around the ice but all bobbing up and down in their skating stride to the same music. While you are watching this, while your gaze might have been elsewhere, one couple will have someone trip, with that couple going down sprawling across the ice, and then you notice couples behind them weaving like crazy not to get caught up in a chain reaction crash.

So there's your model of a non-centralized control close-headway traffic-management system.

Posted by Paul Milenkovic at December 5, 2007 08:02 AM

Aren't these maintenance issues you're concerned about a risk for non-computerized cars? Can't another driver's wheels blow, engine lock up etc and cause a crash?

Posted by Fraser Cain at December 5, 2007 11:55 AM

Can't another driver's wheels blow, engine lock up etc and cause a crash?

Of course they can. Read my post more carefully. I'm concerned about such things happening at high speed, with cars very densely packed (in which there's little margin for failure), which is the goal of the "smart car" initiative. It's much less risky under current highway conditions.

Posted by Rand Simberg at December 5, 2007 12:00 PM

I have just one word for you all. Adaptive Cruise Control.

I understand they have it in the Toyota Avalon, and it uses a TRW millimeter-wave radar. You set a certain speed and if there is a car in front of you, it reduces speed to maintain a following distance. If a car merges in between (I guess "cuts you off" to the hyper-competitive drivers), it adjusts speed and following distance accordingly.

Initially, the driver is ultimately in control to override the system. Just like with regular cruise control -- has the failure of one of those systems ever been blamed in accidents in any serious kind of way. But over time, more features may be added, and you cars may get transponders to report their state to allow the car behind to follow safely. Maybe not the 10-foot separation at first, but increased density over time.

For example, you have a clogged-up road with lots of stop-and-go. Don't you suppose an automatic algorithm could do a better job of following that smooths out some of the velocity variation of the car in front -- it could increase the following distance if the car in front was driving a more variable speed and hitting the brakes more often. One should be able to do a computer simulation if having even a few cars with the smart cruise control could smooth out such traffic so people wouldn't be riding the brakes all of the time.

Posted by Paul Milenkovic at December 5, 2007 08:41 PM

Car computers are pretty sophisticated now. Every month OnStar sends me an email telling me how long I have till I need an oil change and which tires need air and tell me the milage. They do a complete diagnostic on my car and alert me if I have any issues. My mini has run flat tires that I can drive on for 50 miles after the sensor tells me I have a flat. I think the roads would be safer if we had smart cars, but the privacy issue is a major concern.

Posted by Ann NY at December 6, 2007 09:08 AM

And one more thing; I doubt that if the smart car technology were implemented anyone would be allowed to go 90 mph. We recently rented a car in France that had a speed limit of 120 kph. We thought that was just what the top speed of the car was until the day we returned the car, the agency took off the speed restriction and we were allowed to drive back to the rental return as fast as we wanted to.

Posted by Ann NY at December 6, 2007 09:20 AM

I worked on something like this a decade ago (see URL below). Some interesting items:

"Platooning" is the idea of running multiple vehicles together, with very short intervehicle distances within platoons, and long (i.e., car-stopping) distances between platoons. To work it requires a radio link, so that all the cars know when the front car is speeding up or slowing down, rather than waiting for it to "ripple" through the platoon in a wave of dynamic vehicle motion. Platooning can double or even triple the car throughtput. The short spacing can make it *safer* because if a car has a problem, there is not enough distance to build up a significant relative speed before impact. It requires dedicated lanes that only automated vehicles access and use, but is actually much easier to design for than running automated cars mixed with manual traffic.

For automated cars mixed with manual traffic, we can do virtually everything necessary today, and could do most of it back then. The problem is that you need to have automated responses to all the potential "unusual" events (not just 99.9% of the time). Rapid, super-reliable obstacle detection was the most obvious category of "unusual" event. For example, while this project was going on, one day when I drove to work a mylar balloon floated from the side of the road into in front of my car.

My conclusion, BTW, was that cancelling the AHS program meant that we would not get to full automation through the platooning first route, but through the much more technically challenging (but easier deployment transition path) of developing cars that can drive fully automated, mixed with manual traffic. It would probably come from overseas, due to the sorts of legal issues that DocBrown points to that inhibit development in the US. I saw video in the 1990s of a research car driving on the highway in Korea, mixed with ordinary traffic, and with no on in the driver's seat (so no ability to immediately respond to a problem; our cars were driven with drivers in the seat, hands almost on the wheel, even though it turned out the drivers never had to take sudden control). The researcher had cannibalized a key control actuator from a printer...

Tom

http://www.path[dot]berkeley.edu/nahsc/default.htm

([dot] inserted due to the "questionable content" filter)

Posted by Tom at December 6, 2007 01:39 PM

Post a comment