Deception in A Multi-agent Adversarial Game: The Game of Guarding Several Territories

Authors:

Abstract
In this paper, a deceitful behaviour in an adversarial game is investigated. The simulation platform is the game of guarding several territories. In the game, an invader is playing in front of two defenders. A two-level policy system is proposed in this paper. The first level is called lower-level policy, which is the policy to invade or defend a particular territory. The second level is called higher level-policy, which decides about the invader’s behaviour and defenders’ belief. The invader tries deceiving the defenders by repeatedly changing its goal from a fake goal to the true goal. On the other side, the defenders try to guess the goal by the invader’s movement. The lower-level policy is trained via the FACL algorithm, while the higher-level policy is derived via the genetic algorithm. The results show a significant improvement in the invader’s pay-off by behaving deceitfully in the game.

This publication has 8 references indexed in Scilit: