> > 2) Alternatively, we can do a continuous update system, rather than a > > complete refresh. Basically, we keep track of changes on the map, where a > > change is from a path obstacle to a non obstacle, or vice versa.
I guess this itself would be the most significant advance. When no changes happen, we don't need to update the gradient at all. This would even work with current gradient algorithms. By the way: hidden forbidden areas might hit performance really bad, because they mix the fast changing glob movement with the slow changing landscape. Right now as we use highly suboptimal gradient computation this doesn't matter much. But it might ruin our improvement. Especially since the proposed algorithm has a two times as high worst case runtime as the current algorithm. Hidden forbidden areas mix two things that don't belong together. I wanted to do statistics on the ratio of turns in which changes occur to the total number of turns, but I shied away from glob2 code. With this we could estimate the effect of the optimization, which in my opinion will be large. -- Kai Antweiler _______________________________________________ glob2-devel mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/glob2-devel
