blakeelias
commented
6 years ago This is a good idea. Jesse, who I was working with, gave this definition:
Given a shape of volume n find the n/k disjoint extreme points meaning if we recursively break down the bounding box into regions and then find the points that have the greatest Hausdorff distance from any point outside their own region. This defines a set of n/k extreme points for which you find the MST. You then build a one voxel wide skeleton splitting as necessary.
The points resulting from that, we can then connect with a minimum spanning tree, to build out the "skeleton", and then have a few bots fill in on each branch.
Sergey proposed we try to reduce the problem of model construction to a directed graph problem. We need to implement a skeleton detector and then implement a way to walk that graph. We shouldn't forget that vertices of this graph should be weighted as a function of matter that should be created by the robot!