Model for Simultaneous Actions in Situated Multi-agent Systems

Abstract

The main focus of multi-agent research so far has been on concepts and techniques to analyze and specify multi-agent systems. Much less attention has been devoted to the implementation of the concepts and techniques. This paper intends to bridge the gap between the mere concept of simultaneous actions and its implementation. Simultaneous actions are actions that are executed by different agents at the same time. We study simultaneous actions in the context of situated multi-agent systems where agents and objects have an explicit position in the environment. To clarify the concept of simultaneous actions, first we propose a classification for simultaneous actions and illustrate each type with examples. Then we present a generic model for simultaneous actions that is independent of the applied agent architecture. Support for simultaneous actions involves two aspects: first it must enable agents to act together and second, it must ensure that the outcome of a combination of simultaneously performed actions is in accordance with the domain that is modeled. In the model, acting together is established through synchronization, while the domain requirements are ensured through reification of actions and subsequently combining the simultaneously performed actions in accordance with the valid laws. We used the model to implement the Packet–World with centralized as well as with regional synchronization. In the paper we illustrate the model for both approaches and discuss the implications for the complexity of implementation, the autonomy of agents and the scalability of the multi-agent system.