Crash-avoiding cars


I've written before about automatic self-driving cars, both their risks (overregulation due to fear of their use by terrorists) and possible driving forces (oil companies excited by people taking longer trips) and more.

Generally, except for a few specialized applications (such as the automatic parking lot) such cars, if they are to be used where people or cars that may not under network control are present, must start with a basic ability to avoid accidents. In a vigourous debate with friend Charles Merriam last night, the question came up about where the value will lie. Charles is a big proponent of worrying first about crash-avoiding cars.

Right now we all pay from $250 to $500 per year, and often much more, for insurance to cover the risk of accidents. Of course, that's just the financial cost, and financial proxies for suffering, so the real value we would put on an accident resistent car might be much higher. Perhaps $5,000 to $10,000 over the life of the car.

That seems like a highly lucrative market on its own. While the self-driving car has many other long term merits (because you can do other work while moving, and you don't have to park it, and it can appear on demand as a taxi for you) we should be very close to financially justifying the accident-avoiding car today...

In the recent DARPA Grand Challenge, several cars finished the race, travelling 132 miles in 7 hours over rough terrain, avoiding obstacles, including tank traps, and other vehicles. They typically used combinations of cameras and LIDAR to do this, and we can expect their abilities to rapidly improve with time and Moore's law. There was great improvement over the 1st race where nobody got more than a few miles. The challengers of course spent in some cases millions for their technology, but I don't see much in what they did that couldn't become modestly priced if made in millions of units.

The accident-resistence of cars improves if more of them have it, because cars can also network to prevent accidents. An out of control car might spind wildly through a pack of accident avoiding cars, cooperating so that it hits none of them. You will never be able to entirely prevent accidents -- some will be unavoidable due to physics, mechanical failures and wild actions by non-equipped vehicles and crazy drivers -- but one could imagine cutting the accident rate down so insurance is only $50 instead of $500.

Of course, the liability issue must be solved. Today individuals are mostly liable for car crashes. With these advanced cars, software bugs might be liable. Indeed, even though the software is a new force to protect you, it might be found liable for failing to do so, even though the alternative was not to have any protection at all. If every crash is a lawsuit for the vendors, as was the case in civil aviation 20 years ago, things could get hairy. People are much more willing to accept risks where they feel in control. Even if the software goes wrong and kills 1 person while saving 100, people will resist it. Especially if problems lead to deaths that would not have happened without the system.

And some powerful forces might actually fight this. At first, one would imagine an insurance company would cut their rates if you get an accident-avoiding car. But they might eventually realize that full use of this technology puts them out of business, making the auto insurance business as small as the horse and buggy insurance business.



Last night we covered costs, and why the auto insurance costs are not a predictor of the market value. I suppose you wandered off there. See the NHSTA reports; they view the cost as $230 billion for 2004. That would be a reasonable point for collision avoidance, with the market value of auto-driving cars being much higher.

We also talked why networked cars aren't necessary. In detail.

I don't think you were listening. From the past three conversations we have had, I don't think you listen much any more. I'm sorry to lose someone with whom to converse, and I am not sure what happened. I can only hope that your upcoming move will broaden your mind again.

Not sure what's put you in such a mood, Charles, because I'm agreeing with you. I did write that I agreed that the total cost of accidents was much higher than what we pay in insurance. However, the cost of insurance is interesting because people are notoriously bad at factoring in the cost to society, they're even bad at doing the math on the quantifiable costs to themselves. But if the cost of the crash-avoiding car could be clearly offset by savings in insurance, more people would get it and go for it, and more newspaper columns would tell them it's the right choice. Fortunately with cars, people usually get a loan, so they really will compare the added payment on the car loan to the insurance savings. That will be even better with lease payments, because leases factor in the depreciated value of the car at the end of lease time, while car loans pay off the whole car in 4-5 years.

With leases, and a $5,000 crash-avoidance system, leasees would see the lease costing $40 more per month perhaps, and insurance going down by $50 and see it as an easily justified win.

And yes, I also agreed that you don't need to network cars to do crash avoidance and I also agree you need to build the system to work as well as possible without any network information, but that doesn't change that you can in certain circumstances improve things with networking, because other car's sensors will see things your car can't see, and will sometimes give you a chance at a longer time to react. Generally, I have been down on the uses of networking in cars but this is one of the few that has some mild potential.

I've given this some thought previously and was interested in your comments. My conclusion was that autonomous cars are the future as it's virtually the only way you can have a reliable transport system allowing everyone to travel point to point via the optimum route.

There needs to be an incremental change for people to accept it though. Cruise control was the start. Newer cruise control systems can stay a constant distance behind the car infront. I don't know how reliable these systems are but I imagine they are relatively easy to acheive.

I imagine the next step as where your car purely follows the car infront, very closely behind. This would probably be easier than tracking the edges of a road. I accept it would only really be useful on motorways, but it could relatively simply get us most of the way towards autonomous cars and save a lot of accidents. Cars close together will have less air resistance so reduce fuel consumption. A long train of cars driving close together would radically increase the capacity of our current road network and reduce travel times.

The only difficulty is how do you ensure the lead car is trustworthy?

Add new comment