profile image

Speakers

Keynote Speaker I

Dr. Tomas Krajnik

Department of Computer Science, Faculty of Electrical Engineering, Czech Technical University in Prague, Czech Republic


Title: Chronorobotics: Representing the structure of time for service robots

Abstract: Efficient long-term operation of service robots requires to deal with the environment dynamics characteristic for human-populated environments. We will present methods for building spatio-temporal models of environment dynamics from sparse and irregular observations obtained by robots over long time periods. These models provide service robots the ability to predict future states of the environment, leading to gradual improvement of their efficiency. Our presentation will focus on methods for Frequency Map Enhancement (fremen.uk), which represents the environment dynamics in the frequency-spectral domain, and Warped Hypertime (chronorobotics.tk), which represents the dynamics by a set multiple temporal dimensions wrapped into themselves.

Bio: Tom Krajnik (labe.felk.cvut.cz/~tkrajnik) is working in the mobile robotics domain with a particular focus on long-term mobile robot navigation in changing environments. He proposed several methods of representing uncertainty of changing environments leading to gradual improvement of mobile robot efficiency over time. He designed and implemented software libraries for reliable visual navigation and tracking, which were used by the roboticists of the NASA, EPFL, KIT, AIT etc, and contributed to the success of the CTU team during the MBZIRC and DARPA SubT Challenges. He actively cooperates with several research institutes all across the globe and was invited to present his work at world-leading laboratories including CSAIL@MIT, GRASP@UPENN, Oxford or ETH.


Keynote Speaker II

Prof. Robert Babuska

Delft University of Technology, the Netherlands

Title: Reinforcement learning in robot control

Abstract: Reinforcement Learning (RL) algorithms provide a way to optimally solve dynamicdecision-making and control problems. Recent progress in deep learning hasenabled RL to scale to problems that were previously intractable. Notableexamples include complex board games, such as Go, or tasks with high-dimensionvisual inputs, such as video games or robots learning directly from camerainputs. The ability to learn control policies from scratch is an undisputableadvantage of RL, especially for problems where it is difficult or impossible todesign a controller in advance by using model-based approaches. The focus ofthis talk is on the use of reinforcement learning as a tool for feedbackcontrol design in robotics. We will address the aspects of value function andpolicy approximation, using methods ranging from standard basis functionapproximators, through deep neural networks to our new work showing how toincorporate analytical models generated by means of symbolic regression. Thetalk will include examples of problems that can be successfully solved byreinforcement learning as well as by alternative methods and illustrate some ofthe challenges this exciting field of research is currently facing.

Bio: Prof. Robert Babuska received the M.Sc. (Hons.) degree in control engineering from the Czech Technical University in Prague, in 1990, and the Ph.D. (cum laude) degree from TU Delft, the Netherlands, in 1997. He has had faculty appointments with the Czech Technical University in Prague and with the Electrical Engineering Faculty, TU Delft. Currently, he is a full professor of Intelligent Control and Robotics at TU Delft, Faculty 3mE, Department of Cognitive Robotics.In the past, he made seminal contributions also to the field of nonlinear control and identification with the use of fuzzy modeling techniques. His current research interests include reinforcement learning, adaptive and learning robot control, nonlinear system identification and state-estimation. He has been involved in the applications of these techniques in various fields, Homepage: https://www.tudelft.nl/en/staff/r.babuska/.


Keynote Speaker III

Prof. Vaclav Skala

University of West Bohemia, Czech Republic 


Title: Meshfree methods for interpolation and approximation of large and high dimensional data

Abstract: Interpolation and approximation methods are used in many areas. Usually, the definition domain of data is tessellated by triangles or tetrahedrons, etc. Tessellation algorithms for higher dimensions are extremely complex to implement and have very high computational and memory requirements. Therefore, methods based on tessellation are not applicable for large data and higher dimensions.

In this talk, principles of meshless methods will be introduced. These methods lead to a system of linear equations and lead to smooth approximation, in general. The time complexity is nearly independent on the dimensionality of the solved problem. Application of the meshless methods also lead a smooth final approximation.

Nowadays, meshless methods are applied in many areas, e.g. in signal processing, for solution of ordinary and partial differential equations (fluid flow), in interpolation and approximation methods for large data sets, in computer graphics and data visualization, in image processing (inpaiting, image reconstruction) and computer vision, in scalar and vector field data compression., etc.

Bio: Prof. Vaclav Skala as a professor at the University of West Bohemia (UWB), Pilsen [Plzen] at the Department of Computer Science and Engineering. He has been with the Brunel University at London, U.K., Gavle University, Sweden, Moscow Power Engineering Institute, Moscow, Russia and others. He is the Head of the Center of Computer Graphics and Visualization at UWB.

Prof. Skala is a Fellow of the Eurographics Association. He has been serving as an associate editor of prestigious research journals such as Computers and Graphics (Elsevier), The Visual Computer (Springer), Computer Graphics Forum (John Wiley & Sons.) etc. He is the Editor-in-Chief of the Journal of WSCG.

Prof. Skala is active especially within computer graphics and visualization research communities and organizing research oriented conferences, e.g. WSCG, GraVisMa, HCI-Europe etc. He has been also serving as an international program committee member of many international conferences and member of editorial boards.

Prof. Vaclav Skala has published over 120 research indexed papers with more than 598 (Scopus) and 1579 (Scholar) citations.

His current research is targeted to meshless (meshfree) methods for scalar and vector fields approximation, fundamental algorithms and data structures for computer graphics and visualization, geometry algebra application in computer graphics and geometry.


Keynote Speaker IV

Prof. Dr. Reinhard Koch

Christian-Albrechts-University of Kiel, Germany


Title: Depth Cameras for 3D Computer Vision

Abstract: Depth cameras are very useful in 3D computer vision, because they allow to obtain dense depth maps in video-time. Depth cameras usually deliver color (RGB) and depth (D) images, combined into single RGB-D images, with color and depth from the same view point. In my talk I will motivate the use of depth cameras, classity them w.r.t. their operating principles (triangulation and time-of-flight), and will discuss a variety of useful applications that rely on real-time RGB-D Videos. Applications are human-computer interaction and Augmented Reality, Autostereoscopic 3D-Televison, and handling and tracking of deforming objects in real-time.

Bio: Prof. Reinhard Koch received his PhD in 3D Computer Vision from the University of Hannover, Germany, in 1996. He then worked as PostDoc at K.U. Leuven, Belgium, under the guidance of Prof. Luc van Gool from 1996-99, and since 1999 as professor of Computer Vision and Multimedia Information Processing at the Computer Science department of the Christian-Albrechts-University, Kiel, Germany. He has served as dean and is currently vice-dean of the Faculty of Engineering. 

His research interests cover 3D surface reconstruction and object tracking, 3D Computer Vision and Computer Graphics, Light Fields and Augmented Reality, underwater imaging, plankton classification and deep learning techniques for object segmentation and classification. He has published over 250 peer-reviewed papers and is active in the community, currently serving as president of the German Pattern Recognition Society, DAGM, and as German delegate of the International Association of Pattern Recognition, IAPR. See also the homepage http://www.mip.informatik.uni-kiel.de/en