Dynamic Interest Points: A Formalism to Identify Areas to Patrol within a Continuous Environment

Abstract

The multi-agent patrolling problem consists of positioning agents to minimize the idleness, which represents the time difference between two visits of a same location by at least one agent.In the literature, these locations are defined manually by setting static nodes within a graph representation. However, in the context of patrolling a continuous environment, using static nodes cannot guarantee the coverage of the whole environment. In this article, we propose to discretize the continuous environment in order to generate dynamic waypoints called interest points (IP). We prove that these dynamic IP guarantee the coverage of the whole environment while dealing with its topography and the agent's observation range. We evaluated and compared our approach by benchmarking patrolling environment dealing with different observation ranges. Experiments show that dynamic IP locations are adaptive and more efficient to locate high idleness areas compared to static IP approach.