Robust Localization and Efficient Path Planning for Mobile Sensor Networks

Abstract

The area of wireless sensor networks has flourished over the past decade due to advances in micro-electro-mechanical sensors, low power communication and computing protocols, and embedded microprocessors. Recently, there has been a growing interest in mobile sensor networks, along with the development of robotics, and mobile sensor networks have enabled networked sensing system to solve the challenging issues of wireless sensor networks by adding mobility into many different applications of wireless sensor networks. Nonetheless, there are many challenges to be addressed in mobile sensor networks. Among these, the estimation for the exact location is perhaps the most important to obtain high fidelity of the sensory information. Moreover, planning should be required to send the mobile sensors to sensing location considering the region of interest, prior to sensor placements. These are the fundamental problems in realizing mobile sensor networks which is capable of performing monitoring mission in unstructured and dynamic environment.

In this dissertation, we take an advantage of mobility which mobile sensor networks possess and develop localization and path planning algorithms suitable for mobile sensor networks. We also design coverage control strategy using resource-constrained mobile sensors by taking advantages of the proposed path planning method.

The dissertation starts with the localization problem, one of the fundamental issue in mobile sensor networks. Although global positioning system (GPS) can perform relatively accurate localization, it is not feasible in many situations, especially indoor environment and costs a tremendous amount in deploying all robots equipped with GPS sensors. Thus we develop the indoor localization system suitable for mobile sensor networks using inexpensive robot platform. We focus on the technique that relies primarily on the camera sensor. Since it costs less than other sensors, all mobile robots can be easily equipped with cameras. In this dissertation, we demonstrate that the proposed method is suitable for mobile sensor networks requiring an inexpensive off-the-shelf robotic platform, by showing that it provides consistently robust location information for low-cost noisy sensors.

We also focus on another fundamental issue of mobile sensor networks which is a path planning problem in order to deploy mobile sensors in specific locations. Unlike the traditional planning methods, we present an efficient cost-aware planning method suitable for mobile sensor networks by considering the given environment, where it has environmental parameters such as temperature, humidity, chemical concentration, stealthiness and elevation. A global stochastic optimization method is used to improve the efficiency of the sampling based planning algorithm. This dissertation presents the first approach of sampling based planning using global tree extension.

Based on the proposed planning method, we also presents a general framework for modeling a coverage control system consisting of multiple robots with resource constraints suitable for mobile sensor networks. We describe the optimal informative planning methods which deal with maximization problem with constraints using global stochastic optimization method. In addition, we describe how to find trajectories for multiple robots efficiently to estimate the environmental field using information obtained from all robots.