The Gronk
Colon Pounder
This is why I pushed for a navmesh system when the AI replacement was being discussed many moons ago. Although a navmesh utilizes A* for internal calculations it doesn't suffer nearly as much from chess-board movement. Navmeshes are one of the few AI systems (including neural nets) that I haven't had an excuse to code from scratch.Chessboard movement has nothing to do with A*. It has everything to do with the fact that the "graph" is a bunch of cubes. Other pathfinding algorithms would mostly use the same graph. You could mitigate the chessboard effect by connecting the diagonals for example, but there are obvious issues with that: resource usage being the big one. You've mentioned before switching to direct pathing once they get close which may be a good idea but I think there may be some no-trivial and non-obvious problems in practice.
This is all a step beyond where fatal is currently tho. He is necessarily fixing higher level issues (bullet proof glass, looping, etc) before adding refinements.
One thing I think might be worth a try is to add more cost to nodes that lead away from the player. Zombies on the other side of your wall of iron bars should prefer a direct route though the bars to a long jog around the corner to your open door IMO. Similarly pruning (or adding cost) paths that go out of sight would be awesome. Its a bit silly that the zombies "realize" that they can bash though the door behind them and go up the stairs and down a ladder to get at me. Good start on Bandit AI tho
AI is a huge and complicated subject that really requires somebody working on it full time. There's half-a-dozen POI creators and AI is fataal's side project?
On the plus side, I haven't heard any word about behaviour trees being dropped from the plan yet and if they're set up correctly you can create them in xml and parse them at your leisure.
