You could edit a (directed, weighted) navigation graph manually. The A* algorithm isn’t really the best for this, though. Look up Dijkstra’s algorithm. There are several free implementations you could use.

Otherwise you might want tot try implementing some sort of navigation mesh. It describes the area where your entities can walk, so you could cut holes in it where static obstacles are placed, and make sure no navigation mesh border is smaller than the width of your collision geometry.

Hi, I understand the A* algorithm, but I have some trouble doing it in 3D to suit the needs of my RTS

Basically, in the game I’m making, there will be agents with different sizes of OBB collision boxes.

I can use steering behaviours for avoiding other agents, so I don’t need complete dynamic pathfinding. However, there is a problem because different agents have different collision geometry, and structures can be placed in almost any place.

This means that there might be a gap between two structures where some agents can go through and some can’t.

A solution I have found to this problem is to do a sweep of the collision geometry of the agent from start node of the edge the pf algorithm is currently testing, to the end node of that edge. But this is probably a bit overkill since every edge the algorithm tests would also have to create and test with a collision geometry sweep.

What are some reasonable approaches to this problem?