With one accell you can detect all orientation changes except those along the axis aligned with the up/down vector

With two accells you would think things would improve, you do get better precision along two axes and you can detect yaw changes, but now you're still blind to rotations when the line between the two accells is aligned with the up/down vector.

Now with three 3 axial accells there is no blindness to absolute yaw and no gimballock, but i think there is still some hidden issue that would make 4 non-coplanar 3 axial accells necessary ( i once read a little essay on the web on this and i have the impression it explained things all the way to 4 acells)


And that is all with perfect accelerometers, the cheap ones used in mobiles and game controllers have little jitters in the measurement big enough to easilly drift the rotation calculations off the real value fast enough to not be usefull for anything but the briefest usages (even the most expensive ones would still suffer from too much imprecision due to quantum fluctuations and stuff)


Gyrometers (most electronic gyroscopes actually measure rotation rate and not absolute rotation) still suffer from imprecisions and drift, but because the raw values from them are a few steps closer to the result in the math the imprecisions don't get amplified as much and it takes much longer to drift away from the real value significantly.