Prevention of Harmful Behaviors Within Cognitive and Autonomous Agents

Being able to ensure that a multiagent system will not generate undesirable behaviors is essential within the context of critical applications (embedded systems or real-time systems). The emergence of behaviors from the agents interaction can generate situations incompatible with the expected system execution. The standard methods to validate a multiagent system do not prevent the occurrence of undesirable behaviors during its execution in real condition. We propose a complementary approach of dynamic self-monitoring and self-regulation allowing the agents to control their own behavior. This paper goes on to present the automatic generation of selfcontrolled agents. We use the observer approach to verify that the agents behavior respects a set of laws throughout the system execution.