I’ve spend a better part of a day reverse-engineering the “navigation mesh csv” used by PandAI… then I noticed there’s the blender exporter for it written in python, that could probably save me some time, but oh, well, got a script that takes not a mesh but an image and turns that into a PandAI ‘navigation mesh’.
Now you can paint parts of the map that are ‘unwalkable’ and save it in a format understandable by PandAI, so you can have some pathfinding. The map is a small 64x64, I think I may make it twice as big in the future but I need to test how fast the AI works with that kind of ‘navigation mesh’.
I know that PandAI is a bit aggressive and frame rate dependent, but if you run the update not per frame but say every 0.1 seconds with a doMethodLater task and a LerpPosHprInterval to move the actual actor inside the same task, entities can move at the same speed regardless (more or less…) of framerate.
I also had to revert a fix tho the fxaa for nvidia cards, because it didn’t work for the 1.8.1 version as the fixed relied on int shader inputs that where added to later versions (but now I’m supplying the troublesome inputs as explicit floats so I hope it will work).
SCREEN:
SOURCE:
github.com/wezu/koparka
EDIT: changed the default nav-mesh-map-thing to 128x128, it now generates a ~4MB file, but the pathfinding works more or less with the same speed, so I’m gonna leave it at that. Made the script accept any image size if someone wants to use just that part.