The Dynamic Casteship Mechanism for Modeling and Designing Adaptive Agents

It is still a challenge to develop complex multiagent systems with agents that typically exhibit adaptation to various behaviors in different situations. In this paper, we propose an extension of the dynamic casteship mechanism for modeling and designing adaptive agents. In our approach, caste is the modular unit in the design and implementation of multi–agent systems and provides an abstraction of the ways that agents adapt to different behaviors in different situations. The dynamic behaviors of agents are realized as the change of castes that agents bind during execution by “join” and “quit” operations on agent’s casteships. The extended mechanism also enables agents to change the status of its castes to be either active or inactive at run-time. Accordingly, two new operations ‘deactivate’ and ‘activate’ on agent’s casteships are introduced. In the paper, the formal semantics of these operations is rigorously defined. The properties of dynamic binding mechanism are investigated.