BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Computational Optimisation Group - ECPv6.15.11//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
X-WR-CALNAME:Computational Optimisation Group
X-ORIGINAL-URL:https://optimisation.doc.ic.ac.uk
X-WR-CALDESC:Events for Computational Optimisation Group
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-Robots-Tag:noindex
X-PUBLISHED-TTL:PT1H
BEGIN:VTIMEZONE
TZID:UTC
BEGIN:STANDARD
TZOFFSETFROM:+0000
TZOFFSETTO:+0000
TZNAME:UTC
DTSTART:20130101T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTART;TZID=UTC:20141017T150000
DTEND;TZID=UTC:20141017T150000
DTSTAMP:20260406T015756
CREATED:20170124T102139Z
LAST-MODIFIED:20170124T102139Z
UID:567-1413558000-1413558000@optimisation.doc.ic.ac.uk
SUMMARY:Seminar: Bayesian Optimization for Learning Robot Control
DESCRIPTION:Title: Bayesian Optimization for Learning Robot ControlSpeaker: Dr. Marc DeisenrothAffiliation: Department of Computing – Imperial College LondonLocation: Room 217 Huxley BuildingTime: 3:00pm \nAbstract. Statistical machine learning has been a promising direction in control and robotics for more than a decade since data-driven learning allows us to reduce the amount of engineering knowledge that is otherwise required. In real systems\, such as robots\, many experiments can be impractical and time consuming.  I will discuss Bayesian optimization\, an approach to controller learning that is based on efficient global optimization of black-box (utility) functions\, in the context of robot learning. I will demonstrate that this kind of learning is (a) practical and (b) very fast\, i.e.\, it requires only a few experiments\, to learn good controller parameterizations for a bipedal robot. \nAbout the speaker. Marc is PI of the SML group and an Imperial College Junior Research Fellow (tenure-track) with interests in statistical machine learning\, robotics\, control\, time-series analysis\, and signal processing. Marc joined the Department of Computing as a Research Fellow in September 2013. From December 2011 to August 2013 he was a Senior Research Scientist & Group Leader (Learning for Control) at TU Darmstadt (Germany). Marc is still adjunct researcher at TU Darmstadt. From February 2010 to December 2011\, he was a full-time Research Associate at the University of Washington (Seattle). Marc completed his PhD at the Karlsruhe Institute for Technology (Germany) in 2009. He conducted his PhD research at the Max Planck Institute for Biological Cybernetics (2006–2007) and at the University of Cambridge (2007–2009).  Marc’s research interests center around methodologies from modern Bayesian machine learning and their application  autonomous control and robotic systems. Marc’s goal is to increase the level of autonomy in robotic and control systems by modeling and accounting for uncertainty in a principled way. Potential applications include intelligent prostheses\, autonomous robots\, and healthcare assistants.
URL:https://optimisation.doc.ic.ac.uk/event/seminar-bayesian-optimization-for-learning-robot-control/
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=UTC:20141030T140000
DTEND;TZID=UTC:20141030T140000
DTSTAMP:20260406T015756
CREATED:20170124T102139Z
LAST-MODIFIED:20170124T102139Z
UID:566-1414677600-1414677600@optimisation.doc.ic.ac.uk
SUMMARY:Seminar: Visual-Inertial Odometry (VIO) Using Nonlinear Optimization
DESCRIPTION:Title: Visual-Inertial Odometry (VIO) Using Nonlinear OptimizationSpeaker: Dr. Stefan LeuteneggerAffiliation: Dyson Robotics Lab – Imperial College LondonLocation: Room 217 Huxley BuildingTime: 2:00pm \nAbstract. Visual-inertial fusion for state estimation and mapping has recently drawn increased attention. The sensing modalities offer compelling complementary characteristics\, since inertial measurements provide strong short-term temporal correlations\, while visual correspondences in images form spatial (relative pose) correlations. Furthermore\, inertial MEMS sensors have become increasingly small\, cheap\, and accurate. Traditionally\, the visual-inertial odometry problem has been rather addressed with filtering formulations; in this seminar\, however\, an approach using nonlinear optimization is presented — inspired by recent work of the computer vision community solving large reconstruction problems using optimization. The full batch VIO problem becomes untractable quite quickly; mobile robotics\, however\, needs to comply with real-time constraints. To this end\, a framework using partial linearization of error terms along with marginalization (variable elimination) is suggested that allows for a bounded optimization window using the notion of keyframes without compromising the inherent sparsity of the problem. We will go through the necessary mathematical machinery and present a quantitative evaluation as well as qualitative results from on-board Unmanned Aerial Systems (UAS). \nAbout the speaker. Stefan Leutenegger has obtained his PhD from ETH Zurich (Autonomous Systems Lab\, ASL) in 2014\, where he has has worked on solar airplane design from concepts to realization and flight testing\, as well as related algorithms for navigation close to the terrain. His activities covered a broad range from structural\, electrical and software engineering to the development of highly efficient\, robust\, and accurate algorithms for multi-sensor state estimation and mapping.  As part of his PhD work\, Stefan spent three months at the robotics company Willow Garage in Menlo Park\, California\, in 2012 under the supervision of Dr. Kurt Konolige and Dr. Vincent Rabaud. Besides his involvement in engineering and science activities\, Stefan had the ASL-internal lead in the European FP7 Projects “ICARUS” and “SHERPA” since the proposal writing phase. He has furthermore been involved in BSc and MSc student project supervision as well as for teaching a part of ASL’s Master course on Unmanned Aerial Systems. In October 2014\, Stefan started as a lecturer of robotics at Imperial College London\, working in Andy Davison’s Dyson Robotics Laboratory.
URL:https://optimisation.doc.ic.ac.uk/event/seminar-visual-inertial-odometry-vio-using-nonlinear-optimization/
END:VEVENT
END:VCALENDAR