Decentralized area patrolling for teams of UAVs

摘要

In this paper we introduced a novel method for decentralized control of UAVs in patrolling missions. This method is based on social science and is inspired by humans patrolling strategies. In this scenario, all of the UAVs are homogenous and have the same fixed limited communication range. There are some interest points in environment. The desired goal of UAVs is to discover all of interest points as soon as possible. To handle this, we decomposed the area into segments. We did area decomposition in novel and efficiently. Then we set some of UAVs in each subarea. UAVs of each subarea could connect together even under limited communication range. We named these subareas as islands. We introduced new learning algorithm based on probability. In each island, UAVs learn the interest point's appearance pattern. Islands can exchange information together using periodical connection. In this method if some of agents, for any reason get out of work (like battery charging or crash ...), the system will still work properly, because learning algorithm can exchange UAVs between neighbor islands and the recruitment agents do the duties of lost agents. Results show improvement in time and reliability of this method.