Selection of Conflict Resolution Strategies in Dynamically Organized Sensible Agent-Based Systems

Tse-Hsin Liu, K. Suzanne Barber

Abstract: A Multi-Agent System can be seen as a group of entities interacting to achieve individual or collective goals. In the past two decades, researchers have developed various MAS architectures, one being the Sensible Agent (SA) model (Barber 1996). Because one specific level of autonomy is not suitable for all situations in dynamic environments, SAs are equipped with the capability to reason about and switch among levels of autonomy. Typical autonomy levels (which are assigned to goals instead of agents) include: commanddriven, master, consensus, and locally autonomous. The challenge of coordination and conflict resolution in SA-based systems arises from the dynamic organizational structures; when SAs switch their autonomy levels, they also modify their roles and organizational structures. No existing single coordination technique can satisfy such a variety of needs. Results of previous research on various Conflict Resolution (CR) strategies do provide a foundation to solve this problem, but there is limited research focusing on how agents can select proper one. In the work of Adler and his colleagues (Adler, et al. 1989), agents can select one of the following strategies: arbitration, self-modification (independence), centralization, negotiation, priority convention, and mutual accommodation. The criteria agents use to select CR strategies is the network performance. When network traffic is heavy, agents use arbitration for resolving conflicts; when the load is light agents may try negotiation as well as other CR strategies. For dynamically organized Sensible Agents systems, a more advanced decision process is necessary. We propose that a SA should dynamically select a suitable conflict resolution strategy according to: 1) the nature of conflicts (e.g. goal conflict, plan conflict, or belief conflict), 2) the agent’s social roles (represented by its autonomy levels), and 3) its solution preferences (based on an agent’s local view). In addition to using utilities to evaluate potential solutions, agents also use certain indexes to evaluate available CR strategies, and finally agents conduct some trade-off reasoning between solutions and CR strategies. The following simplified formula (may not be linear) shows how an agent can estimate alternative combinations of specific solutions and CR strategies: TotalValue U Utility M Cost CR Cost weight weight weight = × − × − × mod ify CR strategy