Stigmergy for hunter prey problem

Even though Q-Learning is a popular agent learning algorithm, it still has several weaknesses such as slow convergence in large state space. Generalization methods that try to reduce the size of state space may produce a perceptual abasing problem. In this paper we use only two states for a hunter that trics to catch a random or intelligent prey in a 10*10 square domain, thereby reducing the state space while having to face perceptual aliasing. To solve this problem, we investigate the usefulness of stigmergy for hunters in solving the problem of perceptual aliasing when prey is invisible. We show that, stigmergy fairs better than random strategies and multi-step action selection. Furthermore we reduce the size of state space to only one state and by using stigmergy for a blind agent show that it is sufficient for catching the prey in acceptable exploration steps. Finally, we study hunting behavior of several hunter agents without explicit communication among them.