Title: SP-United: Applying Last Action As Feedback For A.I. Agents
Authors: Ching Choo Voon, Ching Siang Boo, Danny Hong Tai Neo, and Liang Jie Wong
Series: Linkping Electronic Articles in Computer and Information Science
ISSN 1401-9841
Issue: Vol. 4 (1999), No. 007/28
URL: http://www.ep.liu.se/ea/cis/1999/007/28/

Abstract: This paper describes the use of a last action feedback loop to compliment the use of a decision tree in a reactive agent. It aims to create a team of reactive agents that is semi - goal orientated.

Computer programs, artificially intelligent or not, will generally display the characteristic of performing identical actions when given near identical situations. This is due to the lack of human intuition to generate the element of unpredictability. The simulator league is also similar. A client in the simulator league no matter how advanced has to make decisions based on its creator's source codes. Since the span of the source codes is usually very general it limits the flexibility of the client player. Hence if we could provide identical situations to the client we would almost be sure that it would react identically. This would be true, especially if its last action was successful. Therefore, if we could just remember this last action taken by the enemy and the condition under which it happened we could exploit this weakness and use it to our advantage.


Original publication 1999-12-15 Postscript Checksum
Checksum (old) Information about recalculation of checksum