Real World Multi-agent Systems: Information Sharing, Coordination and Planning

摘要

Applying multi-agent systems in real world scenarios requires several essential research questions to be answered. Agents have to perceive their environment in order to take useful actions. In a multi-agent system this results in a distributed perception of partial information, which has to be fused. Based on the perceived environment the agents have to plan and coordinate their actions. The relation between action and perception, which forms the basis for planning, can be learned by perceiving the result of an action. In this paper we focus these three major research questions. First, we investigate distributed world models that describe the aspects of the world that are relevant for the problem at hand. Distributed Perception Networks are introduced to fuse observations to obtain robust and efficient situation assessments. Second, we show how coordination graphs can be applied to multi-robot teams to allow for efficient coordination. Third, we present techniques for agent planning in uncertain environments, in which the agent only receives partial information (through its sensors) regarding the true state of environment.