A control structure for agent interaction

This paper describes the interaction control structure of the agents of a traffic-monitoring multiagent system. The goals of an agent are acquired by three mechanisms: agent innate goals; reception of requests in inter-agent communication; and sub-goaling. In contrast with the mainstream view, goals are conditional structures represented by condition-action pairs. We show that condition-action pairs are suitable for representing persistent conditioned goals, nonpersistent conditioned goals, persistent nonconditioned goals, and nonpersistent nonconditioned goals. Knowledge is represented in ACL/SL, the same language used for inter-agent communication. This option eases the process by which request messages generate goals in the procedural memory of the receiver. The evaluation of the conditions of the goals relies on procedural attachment. Each predicate, function and action is attached to specific handlers, which form the interface between the knowledge level of the agent and its internal procedures and data structures