Talk I:Future Mobility: Integrating Data-Driven and Control Methods into Automotive Decision-Making Systems
Time:Monday, 16 December 2019 9:00 a.m. to 11:00 a.m.
Venue: Lecture Hall, 2nd Floor, Zoon A, Advanced Control Base, Nanling Campus
Abstract: Data is everywhere. Modern vehicles are equipped with hundreds of sophisticated sensors that offer necessary information for various functionalities. With vehicle connectivity, these vehicles can be exploited as mobile platforms to crowdsource real-time road and traffic information, which can be utilized to enhance the automotive decision-making systems for improved safety, efficient energy, and ride comfort.
In this talk, I will first present the Vehicle-to-Cloud-to-Vehicle framework and discuss its opportunities and challenges. The focus of the talk will be the exploitation of automotive vehicles to crowd-source road information for collaborative comfort and energy harvesting. I will also talk about some recent work on online driver identification as well as integrating learning and control for efficient system identification and controls.
Talk II:NextGen Modeling and Control: Integrating Real-Time Learning with Control Theory
Time:Tuesday, 17 December 2019 9:00 a.m. to 11:00 a.m.
Venue:School of Artificial Intelligence (Lecture Hall, 6th Floor, Administration Building, Central Campus)
Abstract: While complex engineering systems incorporate first principles based on physical models, they may not make full use of relevant information from real-time data. Exclusively data-driven approaches to complex engineering systems may lead to incorrect and uninformed decisions as they do not incorporate useful information from the engineering and physical models. A hybrid approach that uses real-time data, in conjunction with basic physical and engineering constraints, has the promise to overcome these limitations and can lead to significantly improved decision capabilities.
In the first part of this talk, I will present an online nonlinear system identification algorithm that can simultaneously identify local linear models as well as their validity zones with minimum calibration efforts. It is enabled by integrating an online clustering algorithm with recursive least squares. In the second part of this talk, I will present a safe Reinforcement Learning framework where we learn a safe policy safely by exploiting the system.
BIO：Dr. Zhaojian Li is an Assistant Professor in the Department of Mechanical Engineering at Michigan State University. He obtained M.S. (2013) and Ph.D. (2015) in Aerospace Engineering (flight dynamics and control) at the University of Michigan, Ann Arbor. As an undergraduate, Dr. Li studied at Nanjing University of Aeronautics and Astronautics, Department of Civil Aviation, China. Dr. Li worked as an algorithm engineer at General Motors from January 2016 to July 2017. His research interests include Learning-based Control, Nonlinear and Complex Systems, and Robotics and Automated Vehicles. He is the author of more than 20 top journal articles and several patents. He is currently the Associate Editor for Journal of Evolving Systems, American Control Conference, and ASME Dynamics and Control Conference.
