4 ms·
Just watched the entire video - very interesting; however, there are two major caveats the speaker mentions that is worth noting. 1) This requires expensive pr
by k2xl 10y ago
Just watched the entire video - very interesting; however, there are two major caveats the speaker mentions that is worth noting.
1) This requires expensive precomputation. He mentioned it took 10 hours on one of the large starcraft maps.
2) (I may have misunderstood this point) This does not necessarily respond well to map changes - so in maps where you have the map changing structure (new walls, etc) you will have to run the precomputation again.
With that said, very interesting stuff coming out of Digipen.
- vanderZwan 10y agoI think the Goal Bounding Gate optimisation he talks about could help with both: it's still O(N²), but if N is limited to those gateway nodes that would cut down on costs tremendously. But that all depends on whether it's cheap to find those gateway nodes automatically.