Multi-USV Adaptive Exploration Using Kernel Information and Residual Variance

Abstract

Using a team of robots for estimating scalar environmental fields is an emerging approach. The aim of such an approach is to reduce the mission time for collecting informative data as compared to a single robot. However, increasing the number of robots requires coordination and efficient use of the mission time to provide a good approximation of the scalar field. We suggest an online multi-robot framework m-AdaPP to handle this coordination. We test our framework for estimating a scalar environmental field with no prior information and benchmark the performance via field experiments against conventional approaches such as lawn mower patterns. We demonstrated that our framework is capable of handling a team of robots for estimating a scalar field and outperforms conventional approaches used for approximating water quality parameters. The suggested framework can be used for estimating other scalar functions such as air temperature or vegetative index using land or aerial robots as well. Finally, we show an example use case of our adaptive algorithm in a scientific study for understanding micro-level interactions.