The "input sensors laid out partly at random" is actually a clever trick, which we do in fact use in things we build although not cameras so far as I know. The sensors are in fact arrayed stochastically, which inherently avoids artefacts like aliasing.
Suppose you're looking at, for example a distant checkerboard pattern. If you use a perfect rectangular grid of sensors at certain distances, you will "see" patterns that don't really exist because of the regularity of your sampling. If you instead "randomly" place the sensors according to a rule that limits their proximity you completely avoid this problem.
Pixar's Photorealistic Renderman (and presumably many other modern 3D rendering systems) likewise uses multiple stochastically chosen sample origin points for each pixel, with the same result - no aliasing.