MineSweeper Example
Hello,
I working on a minesweeper type project (ai-junki).
It's mainly for my school presentation.
But i couldn't manage to do "Evolve minesweepers that pick up the mines but avoid another type of object." step. Are there any solution or tips for this ?
Article Link : http://www.ai-junkie.com/ann/evolved/nnt8.html
I tried to supply closest minesweeper with closest mine to system.
I incremented score when the sweeper finds a mine.
I decremented score when the sweeper hits another sweeper.
But it isnt working as intended ... Any suggestion ?
Thanks
November 29, 2005 08:28 AM
It may be that your minesweepers have no way of knowing where other objects are. Trying to get them to avoid other objects when they have no way of knowing where those objects are is like trying to solve a (visual) puzzle blindfolded. The neural network can't be trained to produce outputs that avoid other objects when there are no inputs that give any indication where the other objects are.
Hi,
I am already providing nearest mine and minesweeper to my system.
Also current direction of minesweeper ...
As for output, system returns two variable which are "sweepers left & right tracks" ...
Any idea ?
Do i have to provide all minesweeper locations ?
I am already providing nearest mine and minesweeper to my system.
Also current direction of minesweeper ...
As for output, system returns two variable which are "sweepers left & right tracks" ...
Any idea ?
Do i have to provide all minesweeper locations ?
November 29, 2005 03:36 PM
So if I'm reading the tutorial and what you said correctly, your network currently has 6 inputs. The 6 inputs are 3 vectors of 2 elements each. The vectors are the direction of the minesweeper, the direction to the nearest minesweeper, and the direction to the nearest mine.
Have you tried looking at the inputs that are being generated to verify that they are correct? One possible problem could be that when finding the minesweeper closest to minesweeper A, it is just finding that minesweeper A is the closest instead of excluding it from the candidates.
Are the vectors to the nearest mine and minesweeper unit vectors or are they the actual displacement vectors? If they are just unit vectors, it might not be enough information since there is no way to tell whether the mine or the minesweeper is closer. Maybe try using displacement vectors instead, or add inputs indicating the distances to the nearest mine and minesweeper.
Have you tried looking at the inputs that are being generated to verify that they are correct? One possible problem could be that when finding the minesweeper closest to minesweeper A, it is just finding that minesweeper A is the closest instead of excluding it from the candidates.
Are the vectors to the nearest mine and minesweeper unit vectors or are they the actual displacement vectors? If they are just unit vectors, it might not be enough information since there is no way to tell whether the mine or the minesweeper is closer. Maybe try using displacement vectors instead, or add inputs indicating the distances to the nearest mine and minesweeper.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement