…and by the time that software is so bulletproof, this technology will be 30 years old, too.

That is a bit of an exaggeration, but with modern designs, you need wind tunnel data, you have to program a simulator, so that you can test control software without crashing a zillion expensive hardware systems, you need ejection seats and an array of test pilots trained in a flight simulator that behaves exactly like the hardware is thought to behave, etc.

The advantage here is that you could program the software so that it always stays far, far away from danger zones. With fighter aircraft, that is not an option. However, if you do that, the thing will not sell.

True, but this is also several orders of magnitude simpler than a modern combat aircraft. It has vastly simpler dynamics, it doesn't have to deal with control surfaces, it doesn't have to be stable with irregular patches of supersonic flow on its wings, and so on. It'll probably only take three or six years, I bet.

How does it work - can both rotors be adjusted by computer? In which dimensions?

Just wondering if it would be an interesting project to try to create a drone with two rotors instead of the current crop with four. Might be a nice playground for genetic algorithms.

Then again, of course there are already helicopters. Why do these drones need four rotors anyway? Because it is that much easier to stabilize?

The rotors are fixed. My guess with this is that each fan has a blade running forward and backward underneath it that can be rotated by a computer (thrust vectoring: [1]). You can exert a torque on the vehicle by pushing some of the thrust to the left or right instead of straight down. You can exert a torque forward or backward by adjusting the power going to the forward and backward rotors. So, turning left means vectoring forward thrust to the right and vectoring backward thrust to to the left, giving you a net CCW torque. Sliding left means vectoring both thrusts to the right, accelerating you sideways. Forward and backward are handled by increasing and decreasing power to the fans.

The quadrotor drones are easier to control than helicopters because they don't need any control surfaces at all. You can exert a torque in any direction by adjusting the power to the rotors. Need to go forward? More power to the back, less power in front; the front of the drone dips and the thing drifts forward. Controlling the power to each engine is trivial and the dynamics are easy, so the math is pretty easy to do.

[1]: http://en.wikipedia.org/wiki/Thrust_vectoring