At dinner this evening folks were discussing why the tens of thousands of deaths on U.S. roads every year don’t get people more upset. I said that I was surprised that cars didn’t have transponders and a warning system. For example, a lot of accidents occur as people make left turns or pull out of driveways or side streets onto busy undivided highways. In those cases, a car that was aware of the position and velocity of other cars would be able to suggest “wait” and/or flash a red light in a heads-up display. Airliners have TCAS; why can’t cars have something similar?
[One friend objected that adding advanced electronics of this type would simply enable drivers to pay even less attention than they do currently, thus bringing the accident risk back to where it was (“risk compensation”).]
Brad Templeton discusses this extensively at the robocars url that’s attached.
As far as risk compensation goes, if we can get far enough with robocars, you won’t be _able_ to drive nuttily enough to compensate. Besides, it’s still a big economic gain.
I’d be worried about manipulation. A certain kind of driver would think it funny to increase the power of his transceiver tenfold, making the entire neighborhood light up every time he goes for a drive. This could be prevented with cryptographic distance bounding protocols, but integrating those would not make the system simpler or cheaper.
If the system does cause a lot of false alarms, most drivers will switch it off, if possible, or at least ignore it.
I’m absolutely certain that government regulators would insist that every car broadcasts a unique identifier in their distance warning beacon. Politicians with a security fetish are in love with camera-based number plate readers. Radio ID would make such systems much more reliable, and much cheaper at the same time.
I’m not against the idea, I think most of the issues can be fixed. Just pointing out a few.
I’m afraid your friend’s comment describes what will happen. The logical conclusion is that driving cars should be entirely automated (at least on roads and streets).
Apparently your friend was referring to The Peltzman Effect, which I had never heard of until this morning.
http://www.marginalrevolution.com/marginalrevolution/2010/07/my-entry-1.html
I’ve got a friend who’s worked on some inter-car networked systems, and the problems are hard. Two trivial reasons it’s harder than aviation: you’ve got a lot more vehicles in the space, and there’s a lot of non-vehicle stuff in the environment contributing noise.
And using GPS for position sensing is fine for spaces where less than 500 feet of separation is a big deal and the vehicles are traveling fast enough that you want to be warning at thousands of feet, automobiles routinely travel within 5 feet of each other.
On a related note: Did ABS brakes ever reduce accidents? Or did they just let us drive with a smaller margin for error.
Dan: I didn’t say that putting in transponders and collision-avoidance warning systems was trivial, but hundreds of thousands of deaths and injuries annually is also non-trivial. A $100 GPS can have a database of all of the roads in North America and can determine with high precision the location of a car on one of those roads. There are a lot of mesh networks that work at a range of 1 mile or so. To be useful for avoiding left turn accidents, the system needs to hear from other cars approaching the intersection about their position and velocity. It doesn’t have to be accurate within 5′. In fact, a position report that was accurate to within 50′ would probably be fine. What I am proposing is different from a lane-departure system or “following too close” system (these already exist).
Regarding ABS brakes: http://www.iihs.org/research/qanda/antilock.html says that they have not had a significant effect on accident rates.
“…we can thank the NHTSA for a recent report that at least throws the uncertainty about autonomous safety features into stark relief. The NHTSA had volunteers drive a test track in cars with automatic lane departure correction, and then interviewed the drivers for their impressions. Although the report does not describe the undoubted look of horror on the examiner’s face while interviewing one female, 20-something subject, it does relay the gist of her comments.
After she praised the ability of the car to self-correct when she drifted from her lane, she noted that she would love to have this feature in her own car. Then, after a night of drinking in the city, she would not have to sleep at a friend’s house before returning to her rural home….”
not mine, but I forget where I found it.
Short-term, the paucity of information in road network databases would be a problem. The roads are “center-line-digitized”, and there is no information about roadway width, number of lanes, presence or absence of curbs, presence or absence of traffic signals or stop signs, positions of entrances/exits from parking lots, etc. So internally in a car navigation system, an intersection is a geometric point with no dimension, whereas in reality it is an area of significant size. Although the database providers are moving towards providing such info, it probably won’t be available until some years in the future.
A problem with a GPS-only system would be that GPS does not provide azimuth information, and doesn’t know that a vehicle is turning until it can deduce that from curvature of its path. Solid-state “gyros” and accelerometers can provide that info to a certain extent, but are still not especially cheap.
A more fundamental problem much more resistant to solution is that what you are proposing requires a high degree of “artificial intelligence”. If vehicles A and B are approaching each other from opposite directions on a 2-lane road and pedestrian C steps off the curb such that B needs to swerve momentarily towards A to avoid hitting the pedestrian, what is the system to do? How does it distinguish such a situation from one in which B is starting to erroneously make a left-hand turn that will cause a collision with A?
What action should the system take? If it acts to resist the steering input to prevent the swerve, it may cause B to hit the pedestrian. If it sounds a warning, the distraction may worsen the situation.
Note that one reason why machine translation of languages doesn’t work well in practice is that much language in the real world is imprecise. Often it is even ungrammatical, such that no matter how perfect the parser with respect to the idealized grammar of the source language, it will not be able to properly interpret the input. One of the most difficult aspects of real-world translation is dealing with ambiguities, grammatical errors, and typographical errors. The kind of system you envision would need to cope with analogous challenges.
I like this idea: http://www.cs.toronto.edu/~hehner/Portation.pdf
I worked at Toyota in the 1990s and they were researching such car avoidance systems back then. Fairly sophisticated stuff. Both car-car systems but also intelligent-road systems too. However the liability issues are enormous and frankly humans don’t like change. Also, looking at what happened this year with Toyota (where a number of the claims against Toyota were false) there are few incentives for car companies to come up with new technologies if the liabilities are unknown.
Are you looking for these guys ? http://www.mobileye.com/