Date: October 12 (Tue) 08:30-12:50
Place: 8F Tamra Hall (Track #1), Halla Hall (Track #2)
| Oct. 12 (Tue) | Track #1 (Tamra Hall) | Track #2 (Halla Hall) |
|---|---|---|
| 8:30 – 9:50 | T_11 | T_21 |
| 9:50 – 11:10 | T_12 | T_22 |
| 11:10 – 12:50 | T_13 | T_23 |
- T_11 Gaussian process: Basics and Applications
Prof. Soo Jeon (University of Waterloo, Canada) - T_12 DOB-based Robust Controllers for Linear and Nonlinear Systems (in Korean)
Prof. Juhoon Back (Kwangwoon University) - T_13 Introduction to Multi-agent Systems (in Korean)
Prof. Hongkeun Kim (KoreaTech), Prof. Younghun Lim (Gyeongsang Nat’l University) - T_21 Implementation of MPC (in Korean)
Prof. Jung-Su Kim (SeoulTech) - T_22 Deep Visual-SLAM (in Korean)
Prof. Chang Ho Kang (Kumoh Nat’l Institute of Technology) - T_23 Vision-based Intelligent Surveillance System for Smart City: Concept, Methods, Applications
Dr. Laksono Kurnianggoro (PT Nodeflux Teknologi Indonesia),
Prof. Wahyono Doank (Universitas Gadjah Mada)
- Principle: Prof. Soohee Han (POSTECH) sooheehan@postech.ac.kr
- Vice Principle: Prof. Jung-Su Kim (SeoulTech) jungsu@seoultech.ac.kr
- T_11 Gaussian process: Basics and Applications / Prof. Soo Jeon
- T_12 DOB-based Robust Controllers for Linear and Nonlinear Systems (in Korean)
/ Prof. Juhoon Back - T_13 Introduction to Multi-agent Systems (in Korean) / Prof. Hongkeun Kim, Prof. Younghun Lim
- T_21 Implementation of MPC (in Korean) / Prof. Jung-Su Kim
- T_22 Deep Visual-SLAM (in Korean) / Prof. Chang Ho Kang
- T_23 Vision-based Intelligent Surveillance System for Smart City: Concept, Methods, Applications / Dr. Laksono Kurnianggoro, Prof. Wahyono Doank
Gaussian process (GP) has recently been broadening its applications as a data-driven approach to estimation, machine learning, control and optimization. This mini-lecture will introduce foundations of GP as a non-parametric stochastic process for regression (i.e. learning input-output mappings from empirical data). After understanding basic principles of Bayesian inference with GP, we look at some of its practical considerations. A particular attention will be paid to cases involving sequential data sampling for successive refinement of GP models, which raises questions around sample efficiency (or optimal sample point selection). We will review some of classical (information-theoretic) approaches in sample selection and how they can be blended with domain knowledge for better efficiency. This will help us to understand basic operation of Bayesian optimization (BO) which combines sequential GP regression with careful sample selection.
In practical control systems, it is inevitable to have plant uncertainties and external disturbances, which are major sources to make the robust control problem challenging. Not surprisingly, many researchers have developed various techniques to efficiently cope with the disturbance and plant uncertainties. One of these ideas is to construct an observer or estimator, called disturbance observer, which estimates the disturbance and the effect of uncertainties, and then use this estimate to achieve the control goal. In this talk, disturbance observers for linear and nonlinear systems will be introduced. In addition, applications to nonlinear systems such as quadrotors and robot manipulators will be also addressed.
A multi-agent system is a large scale system that contains multiple (sub)systems.
The multiple systems, called the agents, can interact with each other through a certain communication network. As such, the agents as a group can show rich dynamic behaviors (usually can not be done by a single system). The consensus and synchronization is one of such group behaviors and, over last two decades, has attracted considerable attention due to its wide range of applications. It is basically to reach an agreement among certain variables of interest of the respective agents and is done by exchanging information with neighboring agents. In this tutorial, we introduce the basic theory and applications of the consensus and synchronization. In particular, we review the algebraic graph theory and some fundamental results of consensus and synchronization. We then introduce its possible applications such as formation control, distributed observers, and distributed median computing.
This tutorial lecture introduces the main concepts of model predictive control theory for linear systems and optimization problems for calculating predictive control. A predictive control design technique to ensure the stability of the closed-loop system is described. Numerical methods for simulating predictive control on a PC and implementing it on embedded systems are introduced.
Simultaneous Localization and Mapping (SLAM) aims to build a map of the environment and localize the agent within the environment. It is a critical capability for robotics, especially autonomous vehicles and various deep learning algorithms have more recently been applied to the SLAM problem to improve the performance of localization and mapping. These algorithms are called deep visual SLAM. In this lecture, I introduce new deep learning-based SLAM systems and analyze their structure and performance. Through the analysis and comparison, I finally put forward some open issues and raise some future research directions in this field.
(To be announced.)
