Dealing with norms within the packet world multi-agent simulator
Document typeMaster thesis
Rights accessOpen Access
Norms are expectations of an agent about the behavior of other agents in the society. The human society follows norms such as wait in queues and giving gifts at birthdays. Norms are of interest to researchers because they help to improve the predictability of the society. Multi Agent Systems (MAS) are formed by a set of agents that interact in an environment to engage goals that can be individuals or collective. In these systems norms improve cooperation and coordination between agents and reduce the computation time required by agents to the decision-making process. Also limits the search space since agents shouldn't perform actions that violate the norms of the running environment. Usually it is complicated to de ne norms due to the complexity of the systems and dynamicity of the environment. In Open MAS, agents (devel- oped by di erent companies) compete against each other and it is hard to ensure that all them will behave as expected, this di cult the decision of which norms should be set. Therefore, tools to test and evaluate agent's behavior using norms are needed and a way of doing it is through simulated environments. The purpose of this thesis is to extend the functionalities of the Packet World simulator by implementing a test bed where norms could be easily added and modi ed in order to evaluate the di erent outcomes of these norms. With the Normative Packet World developed, some norms where proposed and evaluated. The layout of this work follows the following structure; in section 2 we give a brief overview of the Packet Work architecture, in section 3 we explain the proposed extension of the system, in section 4 we present the laws already implemented in the Packet World and the institutional norms and norms added, in section 5 we detail the agent's normative behavior developed to test the norms, in section 6 the results of di erent experiments are explained and in section 7 the conclusions of this work are presented.
DegreeMÀSTER UNIVERSITARI EN INTEL·LIGÈNCIA ARTIFICIAL (Pla 2009)