Hi, I''ve done simple maze in 3D with a character inside. I would like to give a certain intelligence to this character. I want him to be able to learn the maze. He should be able after several tries to find the shortest path to the exit. I don''t want to give him directly the solution. He should be able, using a memory system, to find the shortest path to the exit. The first time he gets in the maze, he will have to try several ways before finding the exit. The second time, he will make less mistakes... finally after n tries, he would be able to find the exit using the shortest path. Could you please help me. I''d like to know where i can found papers on this topic. How to organise his memory (data stuctures for the agent memory)... I don''t want to use a algorithm like A* at the beginning in order to find the exit. I know it works but, I want a method with steps...the agent in the maze should learn !! Thank you Eric

Each maze location has a "stain value" that increases every time the character explores it.

When choosing the next step direction, the character tries and points to the "more stained" locations.

As time passes, the "stain value" decreases to avoid bad-paths convergence.

Hope it helps...



[home page] [e-mail]

---
"Lifting shadows off a dream once broken
She can turn a drop of water into an ocean"

Edited by - karmalaa on May 16, 2001 11:39:42 AM

Recursively explore the maze. At each branch point, try each branch. If the branch does not lead to success, your recursion algorithm backtracks to try the other branch. Your agent should remember the better branch to take. Therefore, your agent''s memory would be composed of a the branch areas (as opposed to all areas in the map), and which branch of each branch area is the best.