Dynamic Self-control of autonomous agents
Résumé
Being able to trust in a system behavior is of prime importance, particularly within the context of critical applications as embedded or real-time systems. We want to ensure that a multiagent system has a behavior corresponding to what its developers expect. The use of standard techniques to validate a system does not guarantee it against the occurence of errors in real condition of execution. So, we propose an additional approach of dynamic self-monitoring and self-regulation such that an agent might control, in real condition, its own behavior. Our approach consists in providing the agents with a set of laws that they have to respect throughout their execution. This paper presents a framework which generates agents capable of self-control from an agent model, a behavior description and laws. For that, the framework modifies the agents program by injecting, some checkpoints allowing the detection of particular events. The laws are represented in the agents by Petri nets connected to the checkpoints in order to verify the agreement between their behavior and the laws. The principles of the framework are illustrated on an example.