Minyi Zhong

January 2011
Distributed Control and Optimization in Energy Limited Cooperative Systems
Committee Members: Advisor: Christos Cassandras, SE/ECE; Appointed Chair: Hua Wang, SE/ME; Ioannis Paschalidis, SE/ECE; John Baillieul, SE/ME; Calin Belta, SE/ME

Abstract: Many modern optimization and control tasks can only be accomplished by deploying a distributed cooperative system, which consists of geographically distributed agents working on missions that require their combined efforts, with little or no central coordination. In this dissertation, we first study a typical problem requiring such a setting, the sensor network coverage and data collection mission, where a team of mobile sensors cooperatively monitor their environment and extract information from data sources. We identify three key mission components: coverage control, data source detection and data collection, and propose an end-to-end solution framework.

For coverage control, we develop a gradient-based scheme to maximize the joint detection probability of random events, taking into account the discontinuities introduced by obstacles and limited sensing field of view. The optimization scheme requires only local information at each node and is suitable for distributed implementation. We also propose a modified objective function which allows a more balanced coverage of the mission space when necessary. To facilitate reliable data source detection, we adopt a Bayesian occupancy grid mapping technique to recursively estimate the locations of potential data sources. Once a set of high occupancy probability locations are identified, a dual-objective optimization problem incorporating both coverage and data collection requirements is solved at each node. The interactions among the three components of the sensor network control system are discussed. A simulator and two robotic testbeds are developed to demonstrate our results.

To reduce communication overhead in energy limited distributed cooperative systems, we develop an event-driven communication scheme by focusing on how and when agents should communicate in order to make their information exchange more efficient and thus save energy. We consider the general problem where multiple agents must cooperate to control their individual state so as to optimize a common objective while communicating with each other to exchange updated state information. We obtain conditions under which the optimization process converges with asynchronous communication of state information among agents. We apply this asynchronous (event-driven) approach to the coverage control problem and numerically show that it substantially reduces energy consumption while preserving the same performance as a synchronous algorithm.