Investigate whether constraint aggregation methods may be extended to hierarchical collision detection

[Chad-Peterson]

Hierarchical collision detection can greatly reduce the computational cost of performing what would otherwise be brute-force collision detection. However, performing these hierarchical calculations introduces numerical discontinuities. For instance, the noninterference constraint value might jump as the "interference" changes from comparing two spheres to four.

We already handle jumps in min/max constraints through constraint aggregation, which enables them to stay differentiable. I'm curious if we can blend the CD values in a similar way. For instance, if the first check uses a single hierarchical recursion and the second uses two recursions, can the constraint be treated in an aggregated manner?

[Chad-Peterson]

Also compare constraint aggregation performance against listing all constraints