[gaius]

The article actually nails the real reason: the USAF bought far fewer F22s than it originally planned, so things like this that apply equally to 1 or 1000 have to have their cost spread over far fewer individual units. I'll bet that the F22 software is not actually that much more complicated than that controlling a modern car, yet Ford or whoever can spread the cost over a million units.

[masklinn]

> I'll bet that the F22 software is not actually that much more complicated than that controlling a modern car, yet Ford or whoever can spread the cost over a million units.

I'll bet it is. A car is significantly simpler than a fighter plane (at all levels of resolution), even when you discount weapons and detection systems. Which you can't discount when you're building software for a plane.

Furthermore, the risks involved in a fly-by-wire avionics soft are much greater than in cars in case of bug or crash: there are no drive-by-wire production cars today, although there are many electronics systems.

[edj]

It's true there are no end-to-end 100% drive-by-wire cars. But there are certain drive-by-wire systems in many modern cars. Electronic throttle control[1] is widespread (or universal?) and brake-by-wire[2] is already used by Mercedes and Toyota.

1. http://en.wikipedia.org/wiki/Electronic_throttle_control 2. http://en.wikipedia.org/wiki/Brake-by-wire

[arohner]

Electric steering, too

http://en.wikipedia.org/wiki/Power_steering#Electric_systems

Several years ago, a friend bought a BMW M6 the year they came out. A few days later, he noticed when the tires were pointed straight forward, the steering wheel was not centered. He took it to the shop, and they told them "that's a software bug".

[gaius]

We've had self-landing planes and autopilots for a long time. Self-driving cars are still a long way off. So I would say that don't underestimate what software you need in a car.

And even if the F22 is 10x as complex, Ford still has 100,000 cars to spread the cost across.

[thaumaturgy]

That's the result of a combination of two things: aircraft don't often have one of the major problems that cars have (dense traffic and other obstacles), and that the problem of dissimilar surfaces was solved, for aircraft, by making the surfaces more similar. Cars instead have to be able to negotiate a variety of terrain.

So, safely automating cars is a harder problem, but not because the car itself is a more complex vehicle.

[gaius]

A harder problem requires more complex software, no? The complexity of the vehicle itself is actually not a factor.

[thaumaturgy]

Oh, you're right. Somehow I misread your earlier statement as being about the complexity of the machines themselves.

I agree totally with you then. Even accounting for avionics, control surfaces, target acquisition, and so on and so forth, automating cars is a much harder nut to crack, software-wise.

[bitsai]

Fly-by-wire avionics go beyond "just" self-landing and auto-pilot (and these are no small accomplishments!). In fly-by-wire aircraft, when the flight controls are moved, it's the software that calculates how to move the control surfaces to achieve the desired result. Also, modern fighters are often designed to be inherently unstable, to achieve greater maneuverability. So even in level flight, the software must calculate and order constant tiny movements to maintain stability. These movements happen dozens to hundreds of times a second, and must take into account factors such as current speed, attitude, air flow across the aircraft, etc.