Scalable Sparse Optimization and Distributed Control for Large-scale Autonomy
Time: Wed 2019-12-18 13.15 - 14.00
Lecturer: Dr. Yang Zheng, Harvard University
Bio: Yang Zheng received the B.E. and M.S. degrees from Tsinghua University, Beijing, China, in 2013 and 2015, respectively. He received the DPhil (Ph.D.) degree in Engineering Science from the University of Oxford, UK, in 2019. He is currently a postdoctoral research fellow in the SEAS and CGBC at Harvard University. His research interests include distributed control of dynamical systems over networks, exploiting sparsity in largescale semidefinite programs and sum-of-squares programs, and their applications to autonomous vehicles and traffic systems. Dr. Zheng was a finalist of the Best Student Paper Award at the 2019 European Control Conference. He received the Best Student Paper Award at the 17th IEEE International Conference on Intelligent Transportation Systems in 2014 and the Best Paper Award at the 14th Intelligent Transportation Systems Asia-Pacific Forum in 2015. He was a recipient of the National Scholarship, Outstanding Graduate in Tsinghua University, and the Clarendon Scholarship at the University of Oxford. In 2018, he received the ABTA Doctoral Research Award in Engineering Science.
Abstract: Modern cyber-physical systems, such as transportation systems, the smart-grid, and the internet of things, are of huge scale and have sparse and distributed control logic due to limited information exchange. Many computational and control problems of practical interest remain unsolved due to the issues of complexity and non-convexity. In the first part of this talk, I will focus on two standard optimization problems, known as semidefinite optimization and sum-of-squares (SOS) optimization. By exploiting the properties of chordal graphs and sparse positive semidefinite matrices, I will present decomposition methods that can scale semidefinite and SOS optimization to large-scale instances, achieving massive scalability. The resulting algorithms have been implemented in the open-source solver CDCS. In the second part of this talk, I will present a new distributed control framework centered on the notion of sparsity invariance, which allows addressing the largest known class of distributed control problems with sparsity constraints. The key idea is to approximate a nonconvex sparsity constraint on a product by separate convex subspace constraints on two variables. Besides, sparsity invariance can facilitate solution scalability and model privacy in distributed control of network systems, by combining with chordal decomposition. The final part of this talk will present a real-world application in mixed traffic systems and confirm the significant impact of autonomous vehicles on urban mobility from a control theoretic perspective.