On May 19th 2010, Minyi Zhong defended his PhD dissertation at Boston University. His dissertation topic is “Distributed control and optimization for power limited cooperative systems”. Minyi Zhong is a graduate student of Boston University’s Division of Systems Engineering. His dissertation committee includes Prof. Christos G. Cassandras (adviser) and Prof. John Baillieul, Prof. Yannis Paschalidis and Prof. Calin Belta.
Zhong started with the outline of the talk and a summary of the main contributions of his dissertation. His research work is focused on designing control algorithms for distributed systems, especially those wireless power limited systems such as sensor networks and mobile robotic networks. The application areas of his research include environmental monitoring and exploration, surveillance and reconnaissance, search and rescue in disaster relief, etc.
He then went on to explain that a distributed system consists of a collection of autonomous computing devices, often called nodes or agents, each of which has locally accessible control and data. These nodes communicate over a network through message passing and jointly perform a single task, provide a single service or share some common resources. He pointed out that the main theme of the research in distributed system is the design of intelligence local node algorithms which run on distributed hardware, consume local data and produce some desirable global system-wide behavior.
The main mathematical machinery used in Zhong’s research in distributed control is distributed optimization. In the distributed optimization framework he introduced, a control task of the distributed systems is formulated as a nonlinear optimization problem, which is then decomposed in series of sub-problems and each sub-problem is solved by a node of system. A key issue addressed by Zhong’s research is how to reduce the communication involved in a distributed cooperative system. His solution involves using node state estimation and event-driven communication so that only when a node finds out that its teammate’s estimate of its state is off by a certain threshold, a communication event will be triggered. Two types of threshold process were introduced (static and dynamic) and their convergence properties are analyzed.
He then illustrated the effectiveness of this event-driven communication scheme by applying it to the sensor network coverage and data collection problem, which is another main problem studied in his dissertation. Towards the end of his presentation, he showed talked about the robotic research platforms used in his experiment. The first one is based on Khepera III, which is a miniature robot with onboard processing and communication capability. But it lacks the ability to perform localization without outside help. The next platform is assembled using off-the-shelf components such as iRobot Create, laser range finder, webcam, netbook. Zhong then showed a video of these robots carrying out coverage control mission, in which he demonstrated this platform’s ability to perform self-localization based on laser scan data and vision-based event detection ability. Throughout the defense, Zhong answered questions from the committee members as well as from the general public.
Jiayin Tian