Page 1 of 1

path finding

PostPosted: Wed Dec 29, 2010 11:58 pm
by supagu
so i've done a few experiments with AI, i did a genetic algorithm to get an AI that can avoid obstacles using two feelers. I also just finished AI that can play tic tac toe which uses a neural network and planning. both work well.
So i've decided to try something more complex, I want to make an AI that can navigate an environment. It will need some kind of planning/memory.
Im wondering on how to do this. I could create treat it like my board game (tic tac toe) where the neural network has an input for each grid/tile on the board. I could then input the current location and the destination and the output would be the direction of movement. I'm not really keen on this idea as i think it will take a lot of learning to be able to actually navigate. I wonder is some kind of self organizing map might help speed up the learning process as neurons which are spatially near each other would learn from their neighbors.

alternatively, I can try ray casting to the walls and hope the AI can identify gaps (ie. door ways) and head towards them.

any other ideas?

Re: path finding

PostPosted: Thu Dec 30, 2010 9:02 am
by andrew.kirillov