I wondering if anyone has used or researched LRTA* (or simply just RTA*) and could tell me about the memory complexity of it. From my own research into it(I haven't got around to using it in a project yet) It's extremely high since from what I understand you have to store a table of heuristic values. On a 64 x 64 graph for example (4096 nodes) assuming each heuristic value is represented by float that would mean, if my calculations are correct, you'd need 67 megs of memory per agent. The reason for this is assuming you can travel from one node to any other on the graph, each of the 4096 nodes would need to contain a value for the 4096 other nodes. Is there any way to improve this?

If your graph is fullly connected, yes. If you are doing pathfinding on a 2d map then no. In the LRTA you just need the cost of going through that point. So if you use a box layout, that is four values an element. If you allow diagonals that is 8 values an element. Or if you want to be cheap one per element actually works.

OK, thank you.