This partly happened in tango in the 90's, when Gustavo Naveira, Chicho Frumboli and Fabian Salas came together to create a structural and kinesiologica base for something that previously existed more as "an intuition".

For each basic configuration of leader and follower and their bodies they looked at all permutations.

Then they constrained it first to the subset of those that are possible to do at all, kinesiologically.

Then further to the subset of those that could be danced with reasonable comfort.

And finally the subset of those that are easy enough to be taught to students and would work on a crowded social dancefloor.

None of this was done with the help of computers.

The most systemic documentation of this is possibly Mauricio Castro's book "Tango -- the structure of the dance".

But lately a lot of new books were published on tango technique; I may be out of the loop.

A friend of mine who's also a tango professional is currently looking into the feasibility of doing a PhD thesis on this topic.

He wants to use ML to spit spit out the full motion tree of tango.

Both to be able to document it automatically, i.e. using generated 3D animations, as well as to discover new combinations the manual approach used by Naveira, Frumboli and Salas, over a quarter century ago, will have missed.