Partition Learning for Multiagent Planning

Abstract

Automated surveillance of large geographic areas and target tracking by a team of autonomous agents is a topic that has received significant research and development effort. The standard approach is to decompose this problem into two steps. The first step is target track estimation and the second step is path planning by optimizing directly over target track estimation. This standard approach works well in many scenarios. However, an improved approach is needed for the scenario when general, nonparametric estimation is required, and the number of targets is unknown. The focus of this paper is to present a new approach that inherently handles the task to search for and track an unknown number of targets within a large geographic area. This approach is designed for the case when the search is performed by a team of autonomous agents and target estimation requires general, nonparametric methods. There are consequently very few assumptions made. The only assumption made is that a time-changing target track estimation is available and shared between the agents. This estimation is allowed to be general and nonparametric. Results are provided that compare the performance of this new approach with the standard approach. From these results it is concluded that this new approach improves search and tracking when the number of targets is unknown and target track estimation is general and nonparametric.