عنوان مقاله [English]
In practice, agents of a multi-agent system usually access local information. In a real world example, agents must be able to move through obstacles while the connectivity of the group is maintained. Moving through obstacles needs network flexibility, in which, in an undirected network graph, unnecessary inter-agent connections must be ignored; this needs consensus between agents. In order to reach such a consensus, every agent needs information regarding its neighbors, or, at least, the neighbors of neighbors. Assuming a large scale multi-agent system, the huge amount of information exchange in a network is a serious problem. In this paper, an undirected network graph is presented, which is capable of switching to a directed graph, if needed. Nodes in this graph can perform directional movements, while they only use very local data. Hence, for moving through narrow gates, the network howsconsiderableflexibility. Since, in this method, decisions are only made based on every agents local data, data traffic will not occur, which is a significant point in a practical large scale multi-agent system. It is important to note that in a moving frame with limited communication, agents cannot even recognize the forward direction. Every agent monitors the relative position of its neighbors; network configuration in its neighborhood is the only information available. Recognition of the forward direction depends on the distance and relative position of neighbors. According to the method, if an agent can define one of its neighbors as a leader, it will then show directional movements towards it. It means that the number of other agents and their attraction cannot prevent it from moving towards its leader. In order to prevent total disconnection, an additional algorithm prevents the agent from getting too far away from its following members. Note that every node does not necessarily show directional movement.