Human-in-the-loop vehicle routing policies for dynamic environments

In this paper we design coordination policies for a routing problem requiring human-assisted classification of targets through analysis of information gathered on-site by autonomous vehicles. More precisely, we consider the following problem: targets are generated according to a spatio-temporal Poisson process, uniformly in a region of interest. It is desired to classify targets as friends or foes. In order to enable human operators to classify a target, one of the vehicles needs to travel to the target¿s location and gather sufficient information. In other words, the autonomous vehicles provide access to on-site information, and the human operator provide the judgment capabilities necessary to process such information. The objective of our analysis is to design joint motion coordination and operator scheduling policies that minimize the expected time needed to classify a target after its appearance. In addition, we analyze how the achievable system performance depends on the number of autonomous vehicles and of human operators. We present novel coordination policies between the vehicles and operators and compare the performance of these policies with respect to asymptotic performance bounds.