Mobile sensors management algorithms for environmental monitoring

Access full-text files

Date

2022-05

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Recent progresses in automatic mobile robot platforms make the development of mobile sensors possible. When different kinds of sensors are installed on different mobile agents platforms, we can deploy the mobile sensor system to monitor large scale and dynamically changing environmental fields. Comparing to traditional fixed sensor networks, teams of mobile sensors have the advantage in mobility and flexibility. They are also easier and cheaper than fixed sensor networks to maintain. To deploy mobile sensors to efficiently complete monitoring or tracking tasks, management algorithms are in need. Practical mobile sensor management algorithms should be able to be implemented in real time. Moreover, algorithms with theoretical performance guarantees will have more credits. In this thesis, we investigate in problems related to mobile sensor management algorithms. We start from giving an overview of the current developments of mobile sensor management algorithms. Taking the application of flood monitoring as a concrete example, we first analyze the necessary components in mobile sensor management algorithms. Through this example, we point out that algorithms which can efficiently assign tasks among multi-agent systems and schedule sequences of tasks for mobile sensors are in need. This thesis then study the two topics separately. We first proposed two algorithms that can efficiently assign tasks among multi-agent systems. These two task assignment algorithms can all be implemented in fully decentralized style with reasonable computational complexity. We proved the suboptimality guarantee when the objective function are submodular functions. In the following chapter, we proposed an online scheme to schedule tasks for mobile agents with the goal to seek for extremum values in a field. The proposed online scheme is inspired by the Upper Confidence Bound (UCB) algorithm in bandits problems. In order to allow each agent to compute the UCB efficiently, we proposed a new concept called Dummy Upper Confidence Bound (D-UCB). With a compromise in regret loss, the online scheme can be implemented in real-time using D-UCB. Finally, we consider the case where there are unknown disturbances present in the environment. The online scheme is adapted to detect those disturbances. We theoretically show that both two online schemes will guarantee mobile sensors to track extremum values in the field. All the algorithms proposed in this thesis are validated through numerical experiments. In the end, we show how these proposed algorithms can be combined together to manage mobile sensors efficiently. We also pointed out future directions to continue developing practical mobile sensor management algorithms.

Description

LCSH Subject Headings

Citation