A decision-making architecture for observation and patrolling problems using machine learning

Abstract

Observation and patrolling methods assure the coverage of the entire environment while dealing with moving targets. The efficiency of these methods rely on a wide range of parameters, such as the number of targets, the communication range of the patrolling agent or the map's shape. Thus, in this paper we propose a decision-making tool to optimize a set of parameters among the settings defining the observation and patrolling problem. The obtained optimal configuration has to ensure the expected efficiencies by the user, through the use of evaluation criteria. This tool is based on a simulation-assisted machine learning architecture, which performs a faster prediction response than running the simulation directly to obtain evaluation result. We evaluate the efficiency of the decision-making tool through several scenario, implying one or two parameters to be optimized.