1 PhD defense from ED Mathématiques et Informatique - 1 PhD defense from ED Sciences de la Vie et de la Santé - 1 PhD defense from ED Sciences Physiques et de l'Ingénieur
Université de Bordeaux
ED Mathématiques et Informatique
Optimization of ITS-G5 network resource management to support C-ITS services
by Abdennour RACHEDI (LaBRI - Laboratoire Bordelais de Recherche en Informatique)
The defense will take place at 13h30 - Amphi LaBRI - Laboratoire Bordelais de Recherche en InformatiqueDomaine universitaire, 351, cours de la Libération33405 Talence, France
in front of the jury composed of
- Toufik AHMED - Professor - Université de Bordeaux - Directeur de these
- Gérard CHALHOUB - Professor - Université Clermont Auvergne - Rapporteur
- Thierry GAYRAUD - Professeur - LAAS-CNRS/Université Paul Sabatier (Toulouse III) - Rapporteur
- Mohamed MOSBAH - Professor - Bordeaux INP - ENSEIRB-MATMECA - CoDirecteur de these
- Houda LABIOD - Professeure - Télécom Paristech - Examinateur
Summary
This thesis focuses on resource management in Cooperative Intelligent Transportation Systems (C-ITS) leveraging V2X communications, drawing on emerging technologies such as Artificial Intelligence (AI), network slicing, and Multi-access Edge Computing (MEC). These innovations aim to enhance road safety, traffic efficiency, and network resource optimization.Three primary challenges are addressed. The first pertains to the degradation of communication quality in congested V2X networks, affecting the performance of C-ITS. The second challenge is to ensure low latency for priority services, especially for emergency vehicles requiring reliable communication. The third issue concerns service migration in vehicular networks with MEC, aiming to maintain Quality of Service (QoS) despite high mobility and limited Edge server coverage.The thesis presents three main contributions. The first is a proactive approach for Decentralized Congestion Control (DCC) using LSTM neural networks, optimizing channel performance by predicting Channel Busy Ratio (CBR). This method demonstrated effective resource allocation.The second contribution addresses network slicing in ITS-G5 vehicular communications, with an architecture for the ITS-G5 RAN part and a comprehensive network slicing architecture. This approach aims to maintain adequate performance and security levels for each network slice, confirming the effectiveness of these architectures in traffic management and latency reduction.The last contribution deals with service migration with MEC, formulated as a Markov Decision Process (MDP) and solved using deep reinforcement learning, specifically Deep Q Networks (DQN) and Double Deep Q Networks (DDQN). This strategy balances latency and migration cost, showing that the DDQN method effectively manages migration costs while maintaining optimal QoS levels.In conclusion, this thesis provides robust solutions to the challenges of C-ITS, contributing to significant advancements in road safety, traffic efficiency, and user experience in intelligent mobility.
ED Sciences de la Vie et de la Santé
Integrative study, from the cell to the animal model, of the development of a cell therapy for Parkinson's disease.
by Nicolas PRUDON (Institut des Maladies Neurodégénératives)
The defense will take place at 15h00 - Amphithéâtre du Centre Broca Nouvelle Aquitaine 146 rue Léo Saignat33000 Bordeaux
in front of the jury composed of
- Erwan BEZARD - Directeur de recherche - Université de Bordeaux - Directeur de these
- Philippe HANTRAYE - Directeur de recherche - CEA Paris-Saclay - Examinateur
- András LAKATOS - Directeur de recherche - University of Cambridge - Rapporteur
- Mya SCHIESS - Professeure - University of Texas Health Science Center at Houston (UTHealth) - Rapporteur
Summary
A breadth of preclinical studies is now supporting the rationale of pluripotent stem cell-derived cell replacement therapies to alleviate motor symptoms in Parkinsonian patients. Replacement of the primary dysfunctional cell population in the disease, i.e. the A9 dopaminergic neurons, is the major focus of these therapies. To achieve this, most therapeutical approaches involve grafting single-cell suspensions of DA progenitors. However, a considerable number of cells die during the transplantation process, as cells face anoïkis. One potential solution to address this challenge is to graft solid preparations, i.e. adopting a 3D format. Cryopreserving such format remains a major hurdle and is not exempt from causing delays in the time to effect, as observed with the use of cryopreserved single-cell DA progenitors. The work of this thesis focus on the development of 3D neural microtissues as a cell therapy for PD. The use of a high-throughput cell-encapsulation technology coupled with bioreactors to provide a 3D culture environment enabled the directed differentiation of hiPSCs into neural microtissues. The proper patterning of these neural microtissues into a midbrain identity was confirmed using orthogonal methods including qPCR, RNAseq, flow cytometry and immunofluorescent microscopy. The efficacy of the neural microtissues was demonstrated in a dose-dependent manner in non-clinical studies, using the 6-OHDA-lesioned hemiparkinsonian rat model. The grafts were characterized by post-mortem histological analysis, demonstrating the presence of human dopaminergic neurons projecting into the host striatum. The work reported here is the first bioproduction of a cell therapy for Parkinson's disease in a scalable bioreactor, leading to a full behavioural recovery 16 weeks in the animal model after transplantation using cryopreserved 3D cell format.
ED Sciences Physiques et de l'Ingénieur
Exploiting robot motion to improve situation awareness in human-robot collaboration
by Benjamin CAMBLOR (Laboratoire de l'Intégration du Matériau au Système)
The defense will take place at 10h00 - Ada Lovelace, Centre Inria de l'université de Bordeaux,200 Av. de la Vieille Tour, 33405 Talence
in front of the jury composed of
- Jean Marc SALOTTI - Professeur des universités - Bordeaux INP - Directeur de these
- David DANEY - Directeur de recherche - Centre Inria de l'université de Bordeaux - CoDirecteur de these
- Mickael CAUSSE - Enseignant-Chercheur (ENAC, ISAE) - ISAE/DCAS - Rapporteur
- Anne SPALANZANI - Professeure des universités - Centre Inria de l'Université Grenoble Alpes - Rapporteur
- Aurélie CLODIC - Ingénieure de recherche - LAAS - Examinateur
- Aurélie LANDRY - Maîtresse de conférences - LIP/PC2S Université de Grenoble Alpes - Examinateur
- Hélène SAUZéON - Professeure des universités - Centre Inria de l'université de Bordeaux - Examinateur
Summary
One of the challenges of Industry 4.0 is to preserve the health and comfort of operators while improving their productivity. Collaborative robotics is a solution which, through appropriate assistance, enables the operator to focus on tasks for which he has expertise, while delegating loads and constraints to a collaborative robot. This involves combining the strengths of industrial robots (high physical capacity, repeatability, strength, endurance, speed, etc.) with those of humans (variability, reactions to uncertainty). This research work is part of the ANR Pacbot project (ANR-20-CE10-0005). The overall aim of the project is to design a semi-autonomous cobotic system for assistance, capable of selecting, synchronizing and coordinating tasks distributed between human and robot, adapting to different types of variability in professional gestures, while anticipating dangerous situations. In particular, this system is designed to schedule tasks while minimizing the risk of human error. The study of the human factor in industrial environments has shown that the majority of accidents are due to human error or decision-making. In most cases, the latter are caused by operators poor situation awareness.This thesis proposes to use robot motion as a means of communication that supports human situation awareness in human-robot collaboration. One of the important points of our contribution is that these movements, referred to as signaling motions, can be generated while enabling the robot to perform actions thanks to the redundancy of its joints. The choice of motion as a means of communication was inspired by an analysis of industrial robotics accidents. This analysis highlighted a number of similar accident patterns linked to poor situational awareness. By identifying the risks of human error observed, the choice of using the robot as a communication channel in its own right appears to be a promising accident prevention and safety solution.A beneficial effect of signaling motions on situation awareness has been observed in two experimental studies. In the first study, we showed that they enable a collaborative robot to communicate about its state or actions with its human partners, and thus reintroduce them into the robot's action loop. In the second study, we proposed their use to attract a human's attention and extract him/her from accident-prone situations. We also argued that non-motion could also be considered an effective means of communication. Finally, recommendations were proposed for the design and choice of motion to be generated in different types of context.
