Thesis of Zayed ALSAYED “Characterizing the Robustness and Enhancing the Accuracy of SLAM-based Localization Systems for Autonomous Driving”

ABSTRACT

For autonomous driving, a false estimation of localization could create hazardous situations and threaten lives. Therefore, it is necessary to increase the reliability of localization systems by enhancing their accuracy and defining and detecting their operating limits. This thesis tackles the integration of Simultaneous Localization And Mapping (SLAM) in an autonomous vehicle for outdoor urban and peri-urban environments under real-life conditions.
SLAM offers immediate localization capabilities while enabling simultaneous construction of a map of the surroundings. SLAM does not require any prior knowledge of the environment and is independent of the infrastructures.
The topics this manuscript addresses mainly emerge from the long-term operation, the diversity, the complexity, the dynamicity and the large-scale of the outdoor environments.
The main topics are: 1. robustness of a SLAM solution by detecting its operating limits. 2. accuracy by alleviating the impact of models’ approximations. 3. scalability and resources awareness using a solid map management technique.
Confusing structures in the environment cause SLAM to fail by misleading its estimation process. SLAM failure is a significant issue that should be taken into account in order to build a robust localization system for autonomous driving. Two approaches to detecting situations in which SLAM may fail are proposed.
The first approach constitutes a relevant descriptor vector analyzing solely raw laser data. Hence, it detects a priori potential failure scenarios which makes it independent of the underlying SLAM implementation.
The second approach exploits the likelihood scores distribution, which makes it rely on the estimation process but independent of the sensor used. This approach operates in parallel to SLAM. The decision in both approaches is made using different Machine Learning models. Approximations in SLAM models (e.g. map representation model, displacement model) induce systematic errors in their estimations. To attenuate such errors; our approach uses two types of relevant information: the previous relative pose estimations, and the likelihood scores distribution. The prediction is based on an Ensemble Multilayer Perceptron (EMLP) model to give a proper correction. This correction is applied a posteriori to the SLAM estimation to compensate for the errors.
Moreover, the environment size, which is relatively high, cannot be dictated or limited a priori. Hence, we present a map management technique that is dedicated to 2D gridbased SLAM approaches; such a method ensures seamless navigation with stable resource requirements (i.e. memory and processor load) independently of the size of the environment and the length of the journey.
The approaches presented in this thesis are demonstrated and validated with a series of investigations with an extensive experimental evaluation carried out on open datasets and on our real vehicular platforms under real-life constraints.

VEDECOM is very pleased to announce you the defense of the thesis of Li YU

VEDECOM is very pleased to announce you the defense of the thesis of Li YU

VEDECOM is very pleased to announce you the defense of the thesis of Li YU entitled “Absolute localization by mono-camera for a vehicle in urban environment using Street View.” on Friday 6 april 2018 at Mines ParisTech School.

COMMITTEE
Patrick RIVES, INRIA Sophia Antipolis (Rapporteur)
Paul CHECCHIN, Institut Pascal Université Clermont Auvergne (Rapporteur)
Mme Samia BOUCHAFA, Université d’Évry-Val-d’Essonne (Examinateur)
Fabien MOUTARDE, MINES ParisTech (Directeur de thèse)
Cyril JOLY, MINES ParisTech (Examinateur)
Guillaume BRESSON, Institut VEDECOM (Examinateur)

In a work made at Centre de Robotique and Institut VEDECOM, we studied robust visual urban localization systems for self-driving cars. Obtaining an exact pose from a monocular camera is difficult and cannot be applied to the current autonomous cars. Rather than using approaches like Global Navigation Satellite Systems, Simultaneous Localization And Mapping, and data fusion techniques, we mainly focused on fully leveraging Geographical Information Systems (GIS) to achieve a low-cost, robust, accurate and global urban localization requiring no prior passage of an equipped vehicle and based on a single camera.

Our first task was to design a robotic accessible online database from a dense public GIS, namely Google Maps, which has the advantage to propose a worldwide coverage. We make a compact topometric representation for the dynamic urban environment by extracting four useful data from the GIS, including topologies, geo-coordinates, panoramic Street Views, and associated depth maps. We proposed two localization methods to exploit the GIS: one is a handcrafted features based computer vision approach, the other is a convolutional neural network (convnet) based learning technique.

In computer vision, extracting handcrafted features is a popular way to solve the image based positioning. We take advantage of the abundant sources from Google Maps and benefit from the topo-metric online data structure to build a coarse-to-fine positioning, namely a topological place recognition process and then a metric pose estimation by a graph optimization. The only input of this approach is an image sequence from a monocular camera and the database constructed from Google Maps. Moreover, it is not necessary to establish frame to frame correspondences, nor odometry estimates. The method is tested on an urban environment and demonstrates both sub-meter accuracy and robustness to viewpoint changes, illumination and occlusion. Sparse Street View locations produce a significant error in the metric pose estimation phase. Thus our former framework is refined by synthesizing more artificial Street Views to compensate the sparsity of original Street Views and improve the precision.

However, this method suffers from an important computational time. Since the GIS offers us a global scale geotagged database, it motivates us to regress global localizations from convnet features in an end-to-end manner. The previously constructed online database is still insufficient for a convnet training. We hereby augment the originally constructed database by a thousand factor and take advantage of the transfer learning method to make our convnet regressor converge and have a good performance. In our test, the regressor can also give a global localization of an input camera image in real time.

Articles récents :
VEDECOM meets with local authorities at the Salon des Maires

VEDECOM meets with local authorities at the Salon des Maires

This year once again, VEDECOM and VEDECOM Tech have enjoyed an opportunity to meet with local authority representatives at the Salon des Maires exhibition. The yearly national event for public procurement stakeholders was held on 19-21 November 2019 at the Porte de...

read more

VEDECOM is very pleased to announce you the defense of the thesis of Ali ZIAT, ALI MASRI, Julian GARBISO

VEDECOM is very pleased to announce you the defense of the thesis of Ali ZIAT, ALI MASRI, Julian GARBISO

VEDECOM is very pleased to announce you the defense of the thesis of Ali ZIAT entitled ” learning representation for time serie forecasting and classification” on Monday 16 october 2017 in the Jussieu Campus.
The scientific supervision of the thesis within VEDECOM by Mr. Bertrand LEROY.

COMMITTEE
Ahlame Douzal, Université Joseph Fourier Grenoble 1, Rapporteur
Latiffa Oukhellou, Ifsttar, Rapporteur
Matthieu Cord, Université Pierre et Marie Curie – LIP6, Examinateur
Jean Michel Loubes, Institut de Mathématiques de Toulouse, Examinateur
Bertrand Leroy, Institut VEDECOM, Encadrant
Ludovic Denoyer, Université Pierre et Marie Curie – LIP6, Directeur de thèse
ABSTRACT
In this thesis we develop new methods that address the challenges of time series analysis. Our con-tributions are focused on two tasks: the prediction of time series applied to road traffic prediction and time series classification. Our first contribution presents a prediction and completion methods for multivariate and relational time series. The aim is to be capable of simultaneously predicting the evolution of time series that are linked by a graph (that could represent, for instance, the distance between several sensors), as well as completing the missing values in these series (that can corre-spond for instance to a fault of a sensor during a certain time period). Extensions of this model are proposed and described: first in the context of prediction of heterogeneous time series, and then for predicting time series that have an expressed uncertainty. A model for predicting space-time series is then proposed, in which the relations between different series can be expressed in a more general manner, and where they can be learned.
Finally, we will be interested in the classification of time series. A joint learning model of metrics and classification of time series is proposed and an experimental comparison is performed.

 

VEDECOM is very pleased to announce you the defense of the thesis of Ali MASRI entitled “Multi-Network Integration for an Intelligent Mobility” on Tuesday 28 November 2017 at the University of Versailles-Saint-Quentin-en-Yvelines.
The scientific supervision of the thesis within VEDECOM by Mr. Bertrand LEROY.

COMMITTEE
Mr. Jose Antonio F. de Macedo. Professor: Universidade Federal do Ceará, BRESIL. Reporter
Mr. Thomas Devogele. Professor: Université de Tours, France. Reporter
Mr. Bruno Defude. Professor: Telecom SudParis – Paris Saclay, France. Examiner
Mr. Dimitris Kotzinos. Professor: Université de Cergy-Pontoise, France. Examiner
Mr. Omar Boucelma. Professor: Université de Marseilles, France. Examiner
Mr. Bertrand Leroy. Project Manager: VEDECOM, France. Examiner
Mrs. Karine Zeitouni. Professor: Université de Versailles Saint Quentin- Paris Saclay, France. Director of the thesis
Mrs. Zoubida Kedad. Associate Professor: Université de Versailles Saint Quentin- Paris Saclay, France. Co-director of the thesis
ABSTRACT
Multimodality requires the integration of heterogeneous transportation data and services to construct a broad view of the transportation network. Many new transportation services (e.g. ridesharing, car sharing, bike-sharing) are emerging and gaining a lot of popularity since in some cases they provide better trip solutions. However, these services are still isolated from the existing multimodal solutions and are proposed as alternative plans without being really integrated in the suggested plans. The concept of open data is raising and being adopted by many companies where they publish their data sources to the web in order to gain visibility. The goal of this thesis is to use these data to enable multimodality by constructing an extended transportation network that links these new services to existing ones.    The challenges we face mainly arise from the integration problem in both transportation services and transportation data. Our main contributions are: i) an automatic schema matching approach for geospatial datasets. It uses geospatial web services as mediators to help in automatically matching geospatial properties in geospatial datasets, ii) an approach that enables rich semantic connection generation and allows users to define custom relations between transportation data entities, iii) a multimodal trip planning approach that fully integrates ridesharing solutions within public transportation trip planners.

 

VEDECOM is very pleased to announce you the defense of the thesis of Julian GARBISO entitled ” Fair auto-adaptive clustering for hybrid vehicular networks” on Thursday 30 November 2017 at LINCS / EIT Digital, 23 avenue d’Italie, Paris XIII.
The scientific supervision of the thesis within VEDECOM by Mr. Bertrand LEROY.

COMMITTEE
Reviewers:
Christian BECKER, Professor, University of Mannheim
Ken CHEN, Professeur, Professor, University of Paris XIII
Examiners:
Samir TOHMÉ, Professor, University of Versailles – Saint Quentin
Bertrand LEROY, Project Manager, Vedecom Institute
Invited member: Jeremy PITT, Professor, Imperial College London
Advisors:
Ada DIACONESCU, Tenured assistant professor, Télécom ParisTech
Marceau COUPECHOUX, Tenured assistant professor, Télécom ParisTech
ABSTRACT
For the development of innovative Intelligent Transportation Systems applications, connected vehicles will frequently need to upload and download position-based information to and from servers. These vehicles will be equipped with different Radio Access Technologies (RAT), like cellular and vehicle-to-vehicle (V2V) technologies such as LTE and IEEE 802.11p respectively. Cellular networks can provide internet access almost anywhere, with QoS guarantees. However, accessing these networks has an economic cost.
In this thesis, a multi-hop clustering algorithm is proposed in the aim of reducing the cellular access costs by aggregating information and off-loading data in the V2V network, using the Cluster Head as a single gateway to the cellular network. For the example application of uploading aggregated Floating Car Data, simulation results show that this approach reduce cellular data consumption by more than 80% by reducing the typical redundancy of position-based data in a vehicular network.
There is a threefold contribution: First, an approach that delegates the Cluster Head selection to the cellular base station in order to maximize the cluster size, thus maximizing aggregation. Secondly, a self-adaptation algorithm that dynamically changes the maximum number of hops, addressing the trade-off between cellular access reduction and V2V packet loss. Finally, the incorporation of the theory of distributive justice, for improving fairness over time regarding the distribution of the cost in which Cluster Heads have to incur, thus improving the proposal’s social acceptability.
The proposed algorithms were tested via simulation, and the results show a significant reduction in cellular network usage, a successful adaptation of the number of hops to changes in the vehicular traffic density, and an improvement in fairness metrics, without affecting network performance.

Articles récents :
VEDECOM meets with local authorities at the Salon des Maires

VEDECOM meets with local authorities at the Salon des Maires

This year once again, VEDECOM and VEDECOM Tech have enjoyed an opportunity to meet with local authority representatives at the Salon des Maires exhibition. The yearly national event for public procurement stakeholders was held on 19-21 November 2019 at the Porte de...

read more

VEDECOM présent aux Rencontres de la Mobilité Intelligente à Montrouge

VEDECOM présent aux Rencontres de la Mobilité Intelligente à Montrouge

Les 44èmes Rencontres de la mobilité Intelligente, organisées par ATEC ITS France auront lieu, la semaine prochaine les mardi 24 et mercredi 25 janvier, au Beffroi de Montrouge.

Plus de 90 conférences et tables rondes alimenteront le thème « Quels pilotes pour la mobilité ? » avec 8 grandes thématiques : Vers un paiement plus facile, Le transport de marchandise, Maîtriser et réguler le trafic, Pour une mobilité partagée, Le choix des outils et des stratégies dans les transports, ITS pour l’environnement, Les enjeux du véhicule connecté, Priorité à la sécurité.
Pour VEDECOM cette année pas de stand, ni de présentation de démonstrateur en statique mais une plus forte présence à des conférences ainsi qu’une publication d’une page de pub sur nos offres de formation, dans la revue TEC.

Vous pouvez retrouver le programme en cliquant ici.

Articles récents :
VEDECOM meets with local authorities at the Salon des Maires

VEDECOM meets with local authorities at the Salon des Maires

This year once again, VEDECOM and VEDECOM Tech have enjoyed an opportunity to meet with local authority representatives at the Salon des Maires exhibition. The yearly national event for public procurement stakeholders was held on 19-21 November 2019 at the Porte de...

read more

Belle représentation de VEDECOM à la Conférence Internationale des Transports Intelligents à Rio

Belle représentation de VEDECOM à la Conférence Internationale des Transports Intelligents à Rio

VEDECOM a assisté à la 19ème édition de la Conférence IEEE ITS qui s’est déroulée du 1er au 4 novembre dernier au Windsor Oceanico Hotel à Rio.

Plus plus d’une dizaine d’articles ont été présentés lors de cette conférence par les chercheurs de l’ITE VEDECOM (cf : tableau récapitulatif des présentations).

Cette conférence annuelle est l’une des deux conférences majeures de la IEEE Society on Intelligent Transportation Systems (ITSS).

Cette année, plus de 400 articles ont été présentés sur 4 journées, dans des domaines variés, allant de la perception à la régulation, en passant par l’expérimentation à la régulation.

Au-delà de la valeur scientifique, VEDECOM a pu nouer de nombreux contacts avec des chercheurs et des institutions internationales.

VEDECOM ne manquera pas de transformer l’essai lors de la prochaine conférence IEEE ITSC qui  se déroulera à Yokohama !

Articles récents :
VEDECOM meets with local authorities at the Salon des Maires

VEDECOM meets with local authorities at the Salon des Maires

This year once again, VEDECOM and VEDECOM Tech have enjoyed an opportunity to meet with local authority representatives at the Salon des Maires exhibition. The yearly national event for public procurement stakeholders was held on 19-21 November 2019 at the Porte de...

read more

Atelier VEDECOM le 24 janvier 2017 à Grenoble, sur les méthodes et l’analyse de fouille de donnés, appliquées au véhicule autonome

Atelier VEDECOM le 24 janvier 2017 à Grenoble, sur les méthodes et l’analyse de fouille de donnés, appliquées au véhicule autonome

Article atelier VEDECOM méthodes et l’analyse de fouille de donnés, appliquées au véhicule autonome
Article atelier VEDECOM méthodes et l’analyse de fouille de donnés, appliquées au véhicule autonome
Article atelier VEDECOM méthodes et l’analyse de fouille de donnés, appliquées au véhicule autonome
VEDECOM sera mis à l’honneur à l’occasion de la 17ème conférence « Extraction et Gestion des Connaissances » (EGC) qui se tiendra du 24 au 27 janvier 2017, à Grenoble sur le Campus Universitaire de l’Université Grenoble Alpes.
Cette conférence annuelle réunit des chercheurs et praticiens de disciplines relevant de la science des données et des connaissances. Ces disciplines incluent notamment l’apprentissage automatique, l’ingénierie et la représentation des connaissances, les statistiques et analyses de données, la fouille de données, les systèmes d’information, les bases de données, le web sémantique et les données ouvertes, etc…
VEDECOM organise son workshop et lance un appel à contributions
En amont de cette conférence, des ateliers seront proposés le mardi 23 janvier. C’est dans ce cadre que l’équipe VEH08 de VEDECOM et en particulier Guillaume Bresson, Sébastien Glaser et Mohamed-Cherif Rahal organise un workshop sur les méthodes d’analyse et de fouille de données appliquées au véhicule autonome.

L’objectif de cet atelier est de faire se rencontrer des acteurs de la recherche académique et industrielle afin de débattre sur des techniques d’extraction et de gestion de la connaissance appliquées à des données capteurs (Image, LIDAR, RADAR) qui peuvent être utilisées dans des véhicules a conduite déléguée.

 

L’appel à contribution est ouvert du 1er au 20 novembre 2016.

Pour plus d’informations sur la conférence et les ateliers

Pour consulter le site de l’atelier AFDAV, l’appel à participation ou encore les différentes dates de soumissions spécifiques cliquez ici

Pour en savoir plus sur la conférence EGC rendez-vous sur le site officiel

N’hésitez pas à relayer cet événement auprès de vos équipes ou réseaux.
Vous avez également la possibilité d’y participer en tant qu’auditeur.
Articles récents :
VEDECOM meets with local authorities at the Salon des Maires

VEDECOM meets with local authorities at the Salon des Maires

This year once again, VEDECOM and VEDECOM Tech have enjoyed an opportunity to meet with local authority representatives at the Salon des Maires exhibition. The yearly national event for public procurement stakeholders was held on 19-21 November 2019 at the Porte de...

read more