Category: Uncategorized

IID Seminar: Apprentissage ouvert en robotique
(Open-Ended Learning in Robotics)

Stéphane Doncieux, directeur adjoint
ISIR (Institut des systèmes intelligents et de la robotique)
Sorbonne Université, CNRS, Paris

Wednesday, September 14, 2022, 2:30-3:30 p.m.
Room COP-1168 (Centre d’optique-photonique-laser, Pavillon Vachon)

Abstract

Les robots actuels peuvent accomplir des tâches complexes avec une grande précision, mais pour cela, ils doivent rester dans un environnement contrôlé. Faire face à la variabilité d’un environnement non contrôlé reste un défi. L’apprentissage machine devrait pouvoir donner aux robots les capacités d’adaptation requises, mais la robotique dispose de caractéristiques qui en font un domaine d’application particulièrement exigeant pour les méthodes d’apprentissage. Rendre les robots adaptatifs nécessite de plus de s’intéresser à des apprentissages “ouverts”. Nous positionnerons cette notion par rapport au cadre de l’apprentissage par renforcement et nous présenterons les résultats obtenus dans l’équipe sur les questions que posent un tel apprentissage, que ce soit pour acquérir des représentations d’espaces d’état ou pour explorer dans le cas de récompenses rares. Les applications iront, selon les cas, de problèmes jouets en simulation à la saisie d’objets sur robots réels.

Biography

Stéphane Doncieux is a Computer Science Professor at ISIR (Institute of Intelligent Systems and Robotics), Sorbonne University, CNRS, Paris.
Since January 2019, he has been Deputy Director of ISIR, a multidisciplinary robotics laboratory that brings together researchers in mechatronics, computer science, signal processing and neuroscience. He was coordinator of the DREAM FET H2020 project from 2015 to 2018 (https://dream.isir.upmc.fr/). His research focuses on cognitive robotics and in particular on learning and adaptation with an evolutionary and developmental approach.

The presentation will be given in French and the slides will be in English.

For more information, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca

CeRVIM Webinar: Going beyond 3D to estimate the scene geometry and semantics

Raoul de Charette
Scientifique de recherche, INRIA Paris

Hybrid CeRVIM Webinar (co-modal)
June 14, 2022, 2:00 p.m.
Room (in person) : PLT-1120
Zoom (virtual) : https://ulaval.zoom.us/j/64359268203?pwd=U0lzaGFvNWJTNW1zWDFzbitMbzZCZz09

Abstract
Estimating scene geometry and semantics is a prerequisite for visual systems to interact with our physical 3D world. Because they are largely intertwined, these two cues are better estimated jointly.
In this line of research, I will present some of our recent works that estimate the complete 3D information either leveraging 3D data with lightweight 2D backbones, or using monocular 2D images. Beyond the visible scene parts, we will see that additional insights can boost estimation of the occluded areas.

The presentation will be given in English and the slides will be in English.

REPARTI Workshop 2022: Friday, May 13, 2022, 8h45 – 15h15, on Zoom* within the Acfas Conference.

Three keynote presentations, each followed by a series of short presentations, all of which are offered by researchers and students from the member institutions of REPARTI, will allow you to learn more about the research projects and initiatives carried out by REPARTI members. Here is the detailed program which includes the titles of all of the presentations:

REPARTI Workshop 2022 Program

*To obtain the Zoom meeting web link, please access the Acfas website, https://www.acfas.ca/ and connect to your user account linked to your registration for the Acfas conference. Then go to the Colloque #205 webpage and click on “Accéder à la plateforme”.

CeRVIM Webinar: Initiation à la classification de nuages de points

Geneviève Le Houx
Laboratoire LVSN
Dép. de génie électrique et de génie informatique, Université Laval

May 6, 2022, 11:00 a.m.

Abstract
Ce projet, réalisé dans le cadre du cours GEL-7065 – Lectures dirigées en génie électrique III, porte sur la classification de nuage de points. Les nuages de points proviennent de la base de données “RGB-D Object Dataset” disponible sur le web. Ces données contiennent de l’information sur la position ainsi que sur la couleur de chaque point. Le webinaire présentera les différentes étapes pour effectuer la classification des nuages de points: le prétraitement des données, l’extraction de descripteurs ainsi que l’application de l’apprentissage automatique à des fins de classification.

The presentation will be given in French and the slides will be in French.

To obtain the Zoom meeting web link, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca

CeRVIM Webinar: Lidar Scan Registration Robust to Extreme Motions

Simon-Pierre Deschênes
Norlab (Northern Robotics Laboratory)
Dép. d’informatique et de génie logiciel, Université Laval

April 29, 2022, 1:00 p.m.

Abstract
Simultaneous Localization And Mapping (SLAM) algorithms based on point cloud registration have proven effective in mobile robotics over the last decades. However, they are susceptible to failure when a robot sustains extreme velocities and accelerations. For example, this type of motion can take place after a collision, causing lidar scans to be heavily skewed. While point cloud de-skewing methods have been explored in the past to increase localization and mapping accuracy, these methods still rely on highly accurate odometry systems or ideal navigation conditions. In this presentation, a new point cloud registration algorithm taking into account the uncertainty left after de-skewing a point cloud will be presented and its performance in a SLAM algorithm will be analyzed.

The presentation will be given in English and the slides will be in English.

To obtain the Zoom meeting web link, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca

CeRVIM Webinar: EMT to CT scan image registration of implants used in HDR brachytherapy

Isaac Neri Gomez Sarmiento
Laboratoire LVSN
Dép. de génie électrique et de génie informatique
Université Laval, and
Centre Intégré de Cancérologie – CHU de Québec – Université Laval

April 22, 2022, 11:00 a.m.

Abstract
High dose radiation (HDR) brachytherapy is an internal radiotherapy modality used for local cancer treatment that uses a single HDR seed, that travels inside a patient up to the cancer tumor by means of implants (catheters, needles, applicators). The success of this type of treatment is in part related to the accurate localization of the implants inside the patient by means of 3D medical imaging. 3D coordinates are used for optimizing the delivered dose to the tumor, while sparing surrounding healthy tissues. Currently in clinics, these coordinates can be identified manually or automatically from the 3D medical images but there can be errors when implants overlap, there is not sufficient contrast to identify them or in the case of CT scan, the slice thickness is too high. The goal of this project is to accurately reconstruct these implants with the help of an electromagnetic tracking (EMT) technology and align them in a 3D medical image reference frame (CT scan), using rigid transformation algorithms such as Iterative Closest Point (ICP).

The presentation will be given in English and the slides will be in English.

To obtain the Zoom meeting web link, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca

CeRVIM Webinar: Évaluation en temps réel de la qualité d’un scan 3D du corps humain

Félix-Antoine Demers
Laboratoire LVSN
Dép. de génie électrique et de génie informatique
Université Laval

April 8, 2022, 11:00 a.m.

Abstract

De plus en plus, la numérisation 3D est utilisée dans une variété de domaines, notamment dans le secteur de la fabrication de prothèses. Or, il est souvent difficile pour les utilisateurs d’évaluer la qualité de leur numérisation durant la prise de mesures puisqu’ils ne sont pas spécialisés dans le domaine de la vision artificielle.

L’objectif de ce projet de recherche sera de fournir aux utilisateurs de numérisation 3D une rétroaction en temps réel sur la qualité de leurs mesures en vue de produire un meilleur modèle 3D de la partie du corps étant numérisée.

En utilisant des champs vectoriels comme structure volumétrique, il est possible, à partir de la matrice de covariance présente dans chaque voxel, d’évaluer la variation de surface en ce point. D’autres descripteurs locaux, notamment les normales à chaque point des voxels, permettent d’évaluer la qualité de la numérisation.

L’ensemble de ces calculs peuvent être effectués en temps réel. Ainsi, plusieurs métriques seront utilisées afin de déterminer un système de pointage qui servira de référence pour l’utilisateur pendant qu’il fait la numérisation 3D d’une partie du corps humain.

The presentation will be given in French and the slides will be in French.

To obtain the Zoom meeting web link, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca

CeRVIM Webinar: Mesure de la douleur chez la souris à l’aide de l’IA

William Bonilla
Laboratoire LVSN
Dép. de génie électrique et de génie informatique
Université Laval

March 18, 2022, 11:00 a.m.

Abstract
L’utilisation d’animaux en recherche est un privilège et un moyen de dernier recours. C’est pourquoi, les chercheurs doivent s’assurer du bien-être des sujets utilisés en tout temps. Un moyen d’assurer leur bien-être est de mesurer leur niveau de douleur à l’aide de leur expression faciale. Cette méthodologie est utilisée pour les souris en laboratoire et elle s’appelle le Mouse Grimace Scale (MGS). Ce projet de recherche vise à automatiser ce processus à l’aide de l’apprentissage automatique. La présentation définira les outils utilisés et décrira comment les défis rencontrés ont été surmontés.

The presentation will be given in French and the slides will be in French.

To obtain the Zoom meeting web link, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca

CeRVIM-REPARTI Webinar
Parallel and Cable Robotics Research:
Theoretical and Technological Advancements

Hamid D. Taghirad
Advanced Robotics and Automated Systems Lab (ARAS)
K. N. Toosi University of Technology
Tehran, Iran
Visiting Professor, University of Alberta

Date and Time:
Tuesday, March 8, 2022 (17 Esfand 1400)
12:00-13:00 (Eastern Standard Time: -5:00 GMT) or
20:30-21:30 (Iran Standard Time: +3:30 GMT)

Abstract:
Cable and parallel robotics have been gaining more attention among researchers due to their unique characteristics and applications. Simple structure, high payload capacity, agile movements, and deployable structures are the main characteristics that nominate cable-robots from the other types of manipulators for many applications such as imaging, cranes, agriculture, etc. Interdisciplinary research fields such as dynamic analysis and control synthesis of parallel and cable-driven manipulators by using modern and intelligent approaches will be given further consideration in this presentation.

Kamalolmol® robot is a representative of cable-driven robots developed in ARAS research group, which is a fast deployable edutainment cable-driven robot for calligraphy and painting (chiaroscuro) applications. Additionally, ARAS research exploits the simplicity of cable robots with graph-based optimization and perception algorithms to create commercial inspection and imaging tools for various applications. In this webinar, the underlying concepts of such systems and the current state-of-the-art development of these advancements will be presented.

Bio:

Hamid D. Taghirad has received his B.Sc. degree in mechanical engineering from Sharif University of Technology, Tehran, Iran, in 1989, his M.Sc. in mechanical engineering in 1993, and his Ph.D. in electrical engineering in 1997, both from McGill University, Montreal, Canada. He is currently a visiting Professor, University of Alberta, and professor and director of  Advanced Robotics and Automated System (ARAS), K. N. Toosi University of Technology, Tehran, Iran. He is a senior member of IEEE, and Editorial board of International Journal of Robotics: Theory and Application, and International Journal of Advanced Robotic Systems. His research interest is robust and nonlinear control applied to robotic systems. His publications include five books, and more than 300 papers in peer-reviewed international Journals and conference proceedings.

The presentation will be given in English and the slides will be in English.

To obtain the Zoom meeting web link, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca

CeRVIM Webinar: Online optimization of the imaging parameters of complex super-resolution modalities

Anthony Bilodeau
Centre de recherche CERVO
Université Laval

Feb. 4, 2022, 11:00 a.m.

Abstract
Optical super-resolution fluorescence microscopy is an essential tool in biology to visualize the sub-cellular structures with minimal invasiveness. STimulated Emission Depletion (STED) microscopy allows the nanostructures of biological samples to be investigated, even live, by routinely reaching resolutions below 60nm but is often associated with photobleaching of the fluorescent molecules. Photobleaching can be minimized by the microscopist to a certain extent by careful modulation of the imaging parameters (depletion laser power, excitation laser power, pixel dwell time, and others). This however requires knowledge of the influence of each parameter on the imaging objectives (spatial resolution, photobleaching, signal to noise ratio). More complex imaging schemes for STED microscopy, for example RESCue or DyMIN, were introduced to minimize the impact of the image acquisition on the sample but require more parameters to be carefully calibrated. We thus tackle the online optimization problem of identifying a set of optimal imaging parameters under a multi-armed bandit framework. To facilitate the quantitative validation of our machine learning-based optimization routines for super-resolution microscopy, we developed a STED simulation platform. This platform integrates most imaging parameters and photophysical properties of fluorophores for the simulation of STED microscopy experiments. Preliminary results show that our method can also be transferred to real RESCue and DyMIN experiments by optimizing the imaging parameters which alleviates the required expertise of the microscopist.

The presentation will be given in English and the slides will be in English.

To obtain the Zoom meeting web link, please contact:
Annette.Schwerdtfeger@gel.ulaval.ca