Actually, this "analog" way doesn't work very well. Aside from the problems of not following other constraints (stripes, plaids, etc.), it doesn't handle the problem of "local optima" very well. For example, if a trouser leg gets into one position vs another trouser leg, then it gets "caught", and the pieces in effect get locked together, no matter how much shakin' is goin' on. This is a standard problem with so-called "simulated annealing" (stimulated annealing?!?), where the problem is handled by occasionally raising the "temperature" to high levels to allow for more global rearrangements. But the convergence rate is very slow on combinatorial problems like this. At 03:21 PM 11/2/03 -0800, R. William Gosper wrote:
hgb> I worked on a computer program to virtually "lay out" ("marking") the pattern
pieces on the wool to figure out how much cloth a suit would require. Laying out multiple suits at the same time is typically more efficient in terms of fitting the parts together better,
Mike Sinclair at Microsoft once consulted to an outfit that solved this by analog computation. They laser-cut 1/6 scale patterns from plastic, then shook them for an hour on an inclined table.