I have to implement path finding on GPU as a university project. I am going to use A* (I have used it before for solving 8-puzzles)
However, I can't think of a way to make it efficient for a single agent. I probably have to simulate multiple agents to get reasonable performance speed-up.
I read somewhere that I should spawn a thread for each agent, which seems reasonable.
However, how would the agents avoid each other? The world (2D map) would have to be updated each time one moves (this means that I must have a "central" array where the changes will be written to, which will make performance suffer on a GPU (or not?)). I know this is an AI forum, so the question I am asking is, how can I update each agents position with the least number of updates?