Felipe Meneguzzi

Sensor Placement for Plan Monitoring Using Genetic Programming

Monitoring plan execution is useful in various multi-agent applications, from agent cooperation to norm enforcement. Realistic environments often impose constraints on the capabilities of such monitoring, limiting the amount and coverage of available sensors. In this paper, we consider the problem of sensor placement within an environment to determine whether some behaviour has occurred. Our model is based on the semantics of planning, and we provide a simple formalism for describing sensors and behaviours in such a model. Given the computational complexity of the sensor placement problem, we investigate heuristic techniques for performing sensor placement, demonstrating that such techniques perform well even in complex domains.

Norm Identification in Jason Using a Bayesian Approach

Open multi-agent systems consist of a set of heterogeneous autonomous agents that can enter or leave the system at any time. As they are not necessarily from the same organization, they can have conflicting goals, which can lead them to execute conflicting actions. To prevent these conflicts from negatively impacting the system, a set of expected behaviors – which we refer to as norms – can desirable; to enforce compliance to such norms, sanctioning of violating agents can be used to deter further violations. As new agents enter the system, they must be able to identify existing norms in order to avoid sanctions. In this context, this paper provides two contributions. First, we propose a normative multi-agent system that can be used to evaluate norm-identification algorithms. Second, we validate an existing bayesian norm-identification approach in this system, confirming its positive result in a set of experiments.