If you've been following things, you know that after the great success of the first Darpa Grand Challenge, a new Grand Challenge has been proposed, this time for urban driving. The cars will have to navigate a city with other cars on the road. (I'm going to presume demolition derby style vehicles and speeds.) This time DARPA is providing some funding, though it was impressive how last time the modest (by military standards) $2M prize attained what would have been science fiction just years ago.

So I'm refirming my view that self-driving cars will come to us moderately soon. The technology is very near, and the case is so compelling. In spite of interesting speculations about personal rapid transit, or virtual right-of-way or other items in my transportation category, this is the likely winner because it requires no new infrastructure, and if we let it, it can grow from the ground up.

I'm talking cars that can drive today's roads, and are better at avoiding people and other cars than we are. They do it on their own, though they cooperate where it makes sense to do so but don't have cooperate to work.

The most compelling case is that over 1 million people are killed every year in or by cars, about 42,000 in the USA. In fact, there are over 6 million car crashes reported to police in the USA every year, costing an average of $2,900 per vehicle per year (clearly not all borne by insurance companies.) But if that's not enough, we'll see:

• Self valet parking -- car drives you to front door, then parks itself somewhere cheap.
• Ability to read, work or web surf while in transit
• Dedicated lanes and coordination with timed lights for faster trips.
• Possible eventual ability to reliably go through stop-signs and red lights safely.
• Higher fuel efficiency
• Presumably save hundreds per year on insurace with lower accident rates
• Presumably save even thousands on parking (for CBD commuters.) Parking also possible in cheaper, super-dense remote lots when you do need to park close.
• Car will go to airport to pick up friends.
• Car will run errands to pick up prescriptions and other urgent things. Or people will own or rent small efficient mini-cars to do delivery errands.
• Can't afford a car? Put in a lockbox for your stuff and rent it out as a Taxi when you aren't using it. Or use the cars people are renting out as Taxis.

I would pay double for a car like this, but in fact it's likely to save money, not cost money.

All the other alternatives seem worse. Mass transit is slow at grade and super expensive in tunnels or elevated ROW, and has slow and cumbersome transfers, no personalization and no privacy. PRT requires expensive new ROW. Private driving is of course congested and expensive.

Cost of crashes and traffic update

Let's look at all the costs of crashes and other traffic problems:

• With fatal crashes, of course, the cost of human lives, and suffering for loved ones.
• With injury crashes, the cost of the injury, possibly a lifetime of problems, but also lost work.
• With all crashes, the cost of repairing the cars
• The cost of all the other safety equipment in the cars (though we would probably want to keep most of it unless crashes truly went to an insignificant number.) Still making a car safe in a crash is a large portion of its cost. And we still don't have air bags for the people in the back seat.
• The cost of police, fire and ambulences and other crash-management infrastructure.
• The cost of police to enforce traffic regulations (or the cost of tickets to drivers) and parking regulations.
• For accidents during high traffic times, the cost of traffic delays -- 20 minutes for 3,000 people amounts to 1,000 person hours.
• The need for wider roads to handle human driven traffic, and shoulders for accidents.

Boston Driver

In a recent discussion, the subject of the selfish driver came up. In Boston, driving in traffic is a constant game of chicken. Self-driving cars would of course be programmed to always lose a game of chicken. Done properly, a rogue driver could barrel at full speed into a crowd of self-driving cars and they would, if possible to do safely, part like water around the rogue car. You would actually have to work hard to try and hit one, especially if they are communicating to do this even better. Which brings up the problem, how to deal with the rogue driver, because it now seems the smart thing for that driver to do.

I wrote earlier about the problem of the selfish merge -- a problem we have been unable to solve, where people zoom up to the end in a vanishing lane, causing a traffic jam, because somebody always lets them in, making the zoom-up the fastest strategy. I wondered if a reputation system could help. I don't want to build a system where we track all cars and the rogue driver gets an automatic ticket. Though it would be nice if they did it constantly that perhaps vacant cars would glom around the rogue driver -- reversing the strategy so that they always win a game of chicken instead of always lose -- and pen him in and escort him to the cops.