While robotic cars have very little reason to crash into each other, there will still always be some non-robotic things on the roads -- wildlife and drunken pedestrians for one.

But while some cutback in safety features would be possible, I honestly think that the society will take the other option. Robots have no trouble making driving reasonably safe at much faster speeds than humans can. The advantages to society for raising normal road speeds past 200mph vastly eclipse that of reducing the cost of manufacturing the cars. So I think the future will perhaps see even sturdier cars than today, just to be able to survive hitting a deer (or even a cat) at high speed.

Its not that robots can drive faster, its that they can drive more consistently. I used to drive the infamous Washington DC beltway to work and I can tell you, I'd much rather have a constant 35 than the brief periods of 80 followed but an hour of 7.

When it becomes mostly robots on the road, it will be possible to "program out" a lot of the emergent behaviors of traffic in congested areas.

Very good point. A fascinating study in driver behaviour is to go to a high point where you can observe a congested freeway. The Getty Centre above the 405 Freeway in Los Angeles is an excellent example.

The way I would describe it is watching fast-flowing water. Then block it with a piece of wood, and lift it again, then watch the wave pattern flow down the water.

That's exactly what traffic looks like from above. A couple of slow moving cars, someone doing a stupid lane move, and everyone touches the brakes. This ripples down the flow until, much further down the road, people are all stopped and they have no idea why.

Every road, for a given level of traffic, has an optimum speed that the traffic will flow the best. If robotic control could read that value, and stick to it, you would see Freeways flowing much better.

However, that's all a long way off. I'd settle for widespread fitting of adaptive cruise control and getting people to actually use it.

Keep in mind that the reason for 55mph speed limits is not about safety so much as it is about energy efficiency (set during the oil crises of the 70s, never raised again in many places). Going faster increases energy loss due to drag with the cube of the speed, so I don't think we're going to see 200 MPH speed limits until we get something much more energy efficient.

This could make some of the political problems with upping the speed limit that much go away, though.

Absolutely, but even if you go beyond 55 mph, I think that self-driving cars would still be a net benefit in terms of conserving energy. Traffic jams, which are a tremendous waste of energy, would be eliminated if every car was self-driving.

... something like having every car slipstream the car in front of it?

Yeah, potentially, but you're not going to reach the efficiency of a train as long as you have individual cars with a bunch of different use cases/designs.

...and as long as you have 2500 marginal pounds of vehicle per 1-2 passengers.

Put the largest one in front?

I never opted to take any courses in aerodynamics, unfortunately, but my understanding is that establishing a laminar flow ( http://en.wikipedia.org/wiki/Laminar_flow ) is vital at those kinds of speeds, which requires a surface that the air can flow along smoothly. With a biggest-in-front arrangement, you get some slipstream benefit, but you also get a whole mess of turbulence, and there's no way you're going to rival the speed of a bullet train without a lot more energy spent. You want a mostly smooth/unbroken surface that the air can flow along.

I doubt we'll see road speeds increase any time soon. For one, a car that travels at 200 mph will wear down significantly faster, and get drastically reduced fuel efficiency. Imagine a tire blowout at 200 mph. There are also still human drivers on the road, and it might be difficult to enforce (or even pass) a law that says "human drivers must drive slower than robot drivers".

Excellent point about the non robotic obstacles. I really would just like to add mechanical failure. A perfect robot driver with a working car is one thing, but can that driver handle hydroplaning/tire blowouts/transmission failures?

> A perfect robot driver with a working car is one thing, but can that driver handle hydroplaning/tire blowouts/transmission failures?

Much, much better than any human. Robots fail only once. A single robot being programmed ("getting practice") for a failure case is all robots getting that practice for a failure case. Also, robots are always vigilant and ready for failures, and are capable of doing emergency decisions in fractions of a second, before the humans would likely even know something is wrong.