Autonomous Cooperative Mapping of GPS-Denied Cluttered Environments Using Gaussian Process Regression

Abstract

Multi-agent systems can be used in a range of applications to observe and map spatial-temporal phenomena. In this paper, we have taken the first step to develop a multi-agent environmental monitoring system for fully autonomous exploration and mapping of an unstructured indoor GPS-denied environment. By employing the Gmapping SLAM, the agents cooperatively map a previously unknown environment and explore its entirety. At the same time, the agents are able to successfully map and characterize the temperature distribution inside the room passively using Gaussian Process Regression. The system has been experimentally tested in an indoor cluttered environment, by operation of two Unmanned Ground Vehicles built fully in house. The experimental results show that the proposed system could successfully navigate and explore in the cluttered environment and estimate the spatial distribution of the environment by locating two independent heat sources. It was found that while a passive field prediction approach can approximate the temperature distribution in the room and identify the heat sources, the accuracy of the prediction greatly depends on the proximity of the trajectories that the robots traverse close to the sources.