Automatique du Plateau de Saclay

Séminaire le 12 Juin 2015, 10h00 à SALLE DU CONSEIL DU L2S CENTRALESUPELEC

10h00-11h00  : Florin Stoican, Politehnica University of Bucharest
Title : On corner cutting in multi-obstacle avoidance problems

Abstract : One challenging and not extensively studied issue in obstacle avoidance is the corner cutting problem. Avoidance constraints are usually imposed at the sampling time without regards to the intra-sample behavior of the dynamics. This talk improves upon state of the art by describing a multi-obstacle environment via a hyperplane arrangement, provides a piecewise description of the forbidden regions and represents them into a combined mixed integer and predictive control formulation. Furthermore, over-approximation constraints which reduce to strictly binary conditions are discussed in detail.

11h00-12h00  Julien Marzat, Onera
Title : Model Predictive Control for Cooperative Guidance of Autonomous Vehicles

Abstract : This presentation describes a comprehensive framework for the cooperative guidance of a fleet of autonomous vehicles, relying on Model Predictive Control (MPC). Solutions are provided for many common problems in cooperative control, namely collision and obstacle avoidance, formation flying and area exploration. Cost functions of the MPC strategy are defined to ensure a safe collaboration between the vehicles for these applications. An efficient way to select the optimal cost with limited computation time is also provided. The performance of the proposed approach is illustrated by simulation and experimental results.

We hope, with this survey, to contribute to the preserving of milestones otherwise little remembered in modern texts and, hence, little known to young(er) generations.

Cortical and perceptual processing of visual form

Séminaire le 12 Janvier 2015, 11h00 à en salle Alfred Fessard, CNRS campus de Gif-sur-Yvette, bâtiment 32
Dans le cadre du cycle des conférences Neuro-PSI

Joseph Antony Movshon,  invité par Yves Frégnac, UNIC
Professor at NYU, Director of the Visual Neuroscience Laboratory, Center for Neural Science, New York University

"I am interested in how the brain encodes and decodes visual information, and in the mechanisms that put that information to use in the control of behavior. My research concerns the function and development of the primate visual system, especially the visual areas of the cerebral cortex. My laboratory supports work on neurophysiology, neuroanatomy, psychophysics, and imaging; the main experimental tool is electrophysiological recording from single neurons in monkeys. We stress analytical and quantitative approaches to the study of visual receptive fields. Conceptually, much of this research draws on related work in visual psychophysics, and on computational approaches to understanding brain organization and visual processing."




Covariance Matrix Estimation and Applications in Radar

Séminaire le 8 Janvier 2015, 09h00 à Salle des séminaires, L2S, Supelec

Frédéric Pascal (Supelec/L2S) donnera un Crash Course sur le thème des distributions symétriques elliptiques complexes, des M-estimateurs et de leurs applications en traitement du signal.

Complex elliptically symmetric (CES) distributions have been widely used in various engineering applications where non-Gaussian models are called for. In this tutorial, circularly symmetric CES distributions are surveyed, some new results are derived and their applications e.g., in radar and array signal processing are discussed and illustrated with theoretical and real-word examples, simulations and analysis of real radar data.
A particular emphasis is on maximum likelihood (ML) estimation of the scatter (covariance) matrix parameter.  General conditions for its existence and uniqueness, and for convergence of the iterative fixed point algorithm are discussed in detail. Specific ML-estimators for several CES distributions that are widely used in the signal processing literature are discussed in depth, including the complex t-distribution, the K-distribution, the generalized Gaussian distribution and the recently proposed generalized inverse Gaussian distribution. Also the closely related angular central Gaussian distribution is discussed. A generalization of ML-estimators, the M-estimators of scatter matrix, is also discussed and asymptotic analysis is provided. The tutorial also contains new results on regularized (penalized) M-estimators of scatter, which are needed in low sample support scenarios.
Applications of CES distributions and the adaptive signal processors based on conventional and regularized ML- and M-estimators of the scatter matrix are illustrated in diverse applications such as radar detection problems and in array signal processing applications for Direction-of-Arrival (DOA) and beamforming in low sample support case. Also applications in image processing (e.g., hyperspectral imaging, Polarimetric SAR images) are discussed. Furthermore, experimental validation of the usefulness of CES distributions for modelling real radar data is given.

Frédéric Pascal received the Master's degree ("Probabilities, Statistics and Applications: Signal, Image et Networks") with merit, in Applied Statistics from University Paris VII - Jussieu, Paris, France, in 2003. Then, he received the Ph.D. degree of Signal Processing, from University Paris X - Nanterre, advised by Pr. Philippe Forster: "Detection and Estimation in Compound Gaussian Noise" in 2006. This Ph.D thesis was in collaboration with the French Aerospace Lab (ONERA), Palaiseau, France.
From November 2006 to February 2008, he made a post-doctoral position in the Signal Processing and Information team of the laboratory SATIE (Système et Applications des Technologies de l'Information et de l'Energie), CNRS, Ecole Normale Supérieure, Cachan, France.  Between March 2008 and December 2011 (resp. Jan. 2012 – Dec. 2013), he was an Assistant Professor (resp. Associate Professor) in SONDRA, the French-Singaporean laboratory of SUPELEC. In 2012, he obtained a Research Directorship Habilitation (HDR) thesis in Signal Processing from the University of Paris-Sud. From January 2014, Frederic Pascal is a full Professor in the SONDRA laboratory at SUPELEC. His research interests contain estimation, detection and classification in statistical signal processing and radar processing.


Hyperspectral Unmixing Overview : Geometrical, Statistical, and Sparse Regression-Based Approaches

Séminaire le 11 Décembre 2014, 10h00 à Salle des séminaires L2S Supelec, Gif sur Yvette (91)
Prof. José Bioucas-Dias - University of Lisbon, Portugal

Abstract : Hyperspectral cameras acquire electromagnetic energy scattered in their instantaneous field view in hundreds or thousands of spectral channels with higher spectral resolution than multispectral cameras. This enhanced spectral resolution enables material identification via spectroscopic analysis, which facilitates countless applications that  require identifying materials in scenarios unsuitable for classical spectroscopic analysis. However, due to low spatial resolution of  hyperspectral cameras, microscopic material mixing, and multiple scattering, the spectra measured by these cameras are mixtures of spectra of materials, called endmembers. Thus, accurate estimation requires some sort of spectral separation.

For further informations please click here

Alain Sarlette et Nicolas Boulant

Séminaire le 26 Novembre 2014, 11h45 à Salle du Conseil, L2S, Gif sur Yvette (91)

Details on website :

10h-11h   Alain Sarlette,  INRIA Roquencourt :  Engineering controlled quantum systems
Abstract : we give a short introduction to the emerging field of engineering controlled quantum physical devices. For simplicity we focus on the discrete-time approach and on idealized models that are close to actual experimental realizations. We carefully introduce the basic models involved and two feedback stabilization strategies: one equivalent to classical measurement-based control, and the other working as control by interconnection. We illustrate these strategies on a quantum microwave cavity experiment from the Laboratoire Kastler Brossel at ENS Paris.
Related paper : Stabilization of nonclassical states of one- and two-mode radiation fields by reservoir engineering, Phys.Rev.A (2012)

11h-12h   Nicolas Boulant (UNIRS/NeuroSpin/I2BM/DSV CEA) : Design of non-selective refocusing pulses in magnetic resonance imaging at ultra-high field
Abstract : Ultra-High Field MRI at UHF suffers from severe B1 field inhomogeneities which are detrimental to medical diagnosis. Although many powerful flip-angle homogenizing RF pulse designs have been developed to mitigate this effect, there are still numerous and useful MRI applications where the initial input for an RF pulse can be arbitrary, i.e. not necessarily longitudinal, which means that true rotation matrices need to be designed and must perform well regardless of the input magnetization state. This is a very common scenario in quantum computing applications where control schemes are designed to synthetize logic gates to be effective regardless of the input state. For that purpose, the GRadient Ascent Pulse Engineering (GRAPE) method of Khaneja et al (JMR 2005) developed within the framework of NMR quantum computing has been successfully implemented in our laboratory to design arbitrary refocusing pulses. No approximation is made so that all non-linearities and Larmor frequency offsets are fully taken into account. After introducing the theory, I will present experimental data obtained in-vitro (spin-echo) and in-vivo (SPACE, FLAIR) at 7 Tesla which demonstrates the applicability of the technique as well as the correct control of the nuclear spins

CMAP Ecole Polytechnique : Nov 2014 - Juin 2015

Séminaire le 4 Novembre 2014, 10h00 à Salle de Conférence CMAP 2ème étage aile 00 bâtiment des laboratoires

Consulter en ligne le calendrier du séminaire

Séminaire Luca Zaccarian & Sergio Galeani

Séminaire le 30 Octobre 2014, 10h00 à Salle des conseils - L2S - SUPELEC

LUCA ZACCARIAN, A dynamic input allocation paradigm illustrated by applications
Abstract : Typical control systems designs concentrate on plant output specifications and don't always take sufficiently into account specifications related to its input (e.g., coming from saturation phenomenas). This fact is especially relevant when there are degrees of freedom in the plant input selection due to some kind of acuator  redundancy. In this talk we will present a dynamic input allocation scheme that has been proposed roughly five years ago. The core ideas behind the dynamic scheme will be illustrated while discussing its recent application to: elongation and shape control for Tokamak plasmas, control of parallel hybrid electric vehicles, and satellite attitude control.

Biosketch : Luca Zaccarian received the Ph.D. degree from the University of Roma Tor Vergata, Rome, Italy, in 2000 where he then became first Assistant and then Associate professor until 2011. Since 2011 he is Directeur de Recherche at the LAAS-CNRS, Toulouse, France. Since 2013, he also holds a part-time Associate Professor position at the University of Trento, Italy. His main research interests include analysis and design of nonlinear and hybrid control systems and their applications. Luca Zaccarian is an associate editor of the IEEE Transactions on Automatic Control and Systems and Control Letters, was a member of the IEEE-CSS CEB in 2009-2013, is a member of the EUCA-CEB and served as organizing committee and IPC member for several  international conferences. He was a recipient of the 2001 O. Hugo Schuck Best Paper Award given by the AACC and is a Senior Member of  IEEE.

Sergio GALEANIOn optimal/constrained output regulation for linear fat plants
Abstract : Output regulation is the problem of ensuring that an output is driven to zero despite the influence of an exogenous signal generated by a known system in free evolution (the "exosystem"); such a problem includes as special cases output tracking and disturbance rejection for preassigned classes of references and disturbances. In classic results, when the plant is fat (i.e. it has more inputs than outputs) it is just squared down a priori in order to ensure that a unique, linear regulator exists, thus disregarding the infinitely many available alternatives. In this talk, starting from ideas related to input allocation and focusing on linear fat plants in the presence of  constraints or performance criteria, it is shown that better output regulators can be obtained by scheduling inside the linear variety of linear regulators, and that optimal regulators are nonlinear even in very simple cases.

Biosketch : Sergio Galeani received the PhD degree from the University of Roma Tor Vergata, Rome, Italy, in 2002, where he works as an Assistant professor since 2004. His main research interests include analysis and design of linear and nonlinear control systems ; his more recent work has been focused on hybrid output regulation and on nonlinear optimal output regulation for linear fat systems. Sergio Galeani is a member of the IEEE-CSS CEB since 2010, and served as organizing committee and IPC  member for several international conferences.


General factorization approach to frequency domain robust regulation

Séminaire le 10 Juin 2014, 14h30 à Supélec, Gif sur Yvette, Room C.3.12
Petteri Laakkonen (Université de Tampere, Finlande)

Vidyasagar and others developed frequency domain robust regulation theory for rational transfer matrices around 1980s. This theory is based on coprime factorizations and allows parameterization of all robustly regulating controllers and simple characterization of the famous internal model principle - every robustly regulating controller contains a suitably redublicated model of the dynamics to be controlled. These results of rational matrices have been generalized to algebraic structures suitable for infinite-dimensional systems. The problematic part is that coprime factorizations do not necessarily exist or they may be hard to find in algebraic structures for infinite-dimensional systems.

The purpose of this talk is to present some new results on frequency domain robust regulation that use only a left or a right coprime factorization or non-coprime factorizations. We first introduce the basic results for rational functions using coprime factorizations and finally proceed to an abstract algebraic setting where we develop robust regulation theory without using coprime factorizations.

Advanced Techniques of Radar Detection in Non-Gaussian Background

Séminaire le 11 Juin 2014, 10h00 à Supelec, Gif/Yvette - Salle du Conseil du LSS, niveau B4
Dr Maria S. Greco - University of Pisa, Italy

In modern high resolution radar systems, many statistical analysis have proved that the typical radar disturbance, the clutter, is not Gaussian-distributed. In this background the detectors designed for working in Gaussian clutter suffer a huge increase of the false alarms and losses in the performance. For this reason in the last 25 years much attention has been devoted to the modeling of the clutter echoes and to the design of optimum and suboptimum detectors for target detection in non-Gaussian background. Aim of this talk is to describe the state-of-the-art approaches to the coherent target detection in such background.

After a brief description of the modern statistical and spectral models for high resolution sea and ground clutter, with particular attention to the popular compound-Gaussian model, the talk will focus on the design of coherent optimum and suboptimum detectors and to their performance derivation and analysis against such compound-Gaussian background. Different interpretations of the various detectors are provided to highlight the relationships and the differences among them.

The talk will be completed by the analysis of the adaptivity in the detector design, with particular attention to the disturbance covariance matrix estimation. Some new results on naive, robust and fully-adaptive estimation approaches will be provided together with some hints on possible future research directions. 

"Differential Algebra : Applications, Software and Theory"

Séminaire le 27 Mai 2014, 14h30 à Supélec, Gif sur Yvette, room C.3.12
François Boulier

Talk by François Boulier, Université de Lille I, LIFL.

The differential algebra of Ritt and Kolchin is an algebraic theory for systems of nonlinear differential equations. It has been a very active research topic of the Computer Algebra Group of Lille for 20 years, with a focus on algorithms, software and applications. 

In this talk, I will present an introduction to the mathematical theory and some of our results: an open source library which can be used through the MAPLE DifferentialAlgebra package, a recent algorithm which is useful for converting differential to integral equations and is thus important for parameter estimation problems (joint work with Inria/Non-A), ...
Link to seminar page