fil / attitude Goto Github PK

instead of {attitude}

As noted by @jrus there are many ways to encode the angle. The useful functions f(a) are:

• tan(a/4): stereographic, ‘Modified Rodrigues Parameters’.
• tan(a/2): gnomonic, ‘Rodrigues Parameters’, ‘Gibbs vector’.
• a: equidistant, logarithm vector.
• (vector part of the) unit quaternion: Euler angles.

The current code uses f(a) = a ; we have to settle on a default.

The stereographic option seems the best: it allows to use the vector representation to do optimization (via gradient descent for example), and maybe to compute the average of an array of rotations (not sure about this—maybe the equidistant is better?).

See d3/d3-geo#194 (comment)

Using https://observablehq.com/@fil/sinpi-cospi will probably be better than the current "exact cos" strategy. (The idea is to have an exact rotation matrix when angles are multiples of 90°.)