|
|
|
|
|
|
|
|
|
Alix Agnès
Dates : Du 06/05/24 au 25/08/2024
Laboratoire ou Entreprise : Karlsruh Institute of Technology
Confidentiel : Non
Lieu : Postfach 3640, Karlsruhe, Allemagne
Titre du stage : A tabletop social robot to motivate office workers to walk and stretch
Encadrant : Barbara Bruno
Résumé : The goal of this internship at the Karlsruhe Institute of Technology in Germany was to
design, build, and program a tabletop social robot tasked with seeing if their user is sitting
down too long and if that is the case, motivate them to get up and take a break for their
health. The first part of the internship was spent analysing the potential users of the robot
and their needs, and establishing the general design for the robot’s shape. Next, a significant
part of the internship was spent designing the different parts of the robot, printing these
parts with a 3D printer, and then assembling the robot’s body entirely. Finally, the last
part of the internship was spent working on the electrical components of the robot : it is
controlled by a Raspberry Pi 4 that actions the servomotors making the different body
pieces move. A the end of the internship, the robot is fully built and has a set of positions
pre-programmed that indicate whether or not the user should take a break. To make the
robot operational, these positions have to be linked to a motion detecting camera : once
this is done, the robot has to be tested on a significant enough sample of people to confirm
its efficiency.
Jury : Andreas Rauh and Luc Jaulin
Romain Bornier
Dates : Du 01/05/24 au 01/09/24
Laboratoire: Tokyo University of Science (TUS)
Confidentiel : Non
Lieu : 6-3-1 Niijukuku, Katsushika-ku, Tokyo, Japon
Titre du stage : Control of humanoid robot using Machine Learning
Encadrant : Eiichi Yoshida
Résumé : This project is part of a four-month international internship at Professor Yoshida’s
laboratory at the Tokyo University of Science, under the supervision of Dr. Marwan Hamze.
The main objective was to contribute to research on the application of reinforcement
learning for the control of robotic arms.
My work involved developing and implementing control algorithms for a robotic arm,
both in simulation and in real-world environments. Reinforcement learning allows for
the avoidance of complex kinematic models, enabling the robot to optimize its behavior
through direct interaction with its environment.
I focused on optimizing the control of the xArm6 robotic arm by adapting methods from
the scientific literature. I initially tested these algorithms in simulation before applying
them in real-world environments to assess their robustness. My goal was to acquire skills
in reinforcement learning applied to the control of humanoid robots, with the aim of
controlling the Kawasaki Kaleido robot, which measures 1.80 m and weighs 80 kg.
This project enabled me to enhance my technical skills in robotics and artificial
intelligence while contributing to applied research in this rapidly growing field.
Jury : Luc Jaulin, Marit Lahme
Tiphaine C.
Dates : Du 01/05/24 au 21/08/2024
Laboratoire : Crossing - IRL CNRS 2010
Confidentiel : Non
Lieu : The University of Adelaide, North Terrace Campus, Adelaide, Australie
Titre du stage : Développement d'un simulateur Python de suivi du RIPAM et d'un outil de suivi de l'AIS des navires.
Encadrant : Benoît Clément
Résumé : In maritime navigation, autonomous vessels have emerged as a prominent and widely accepted
solution. However, achieving full autonomy for marine vessels requires the development of robust
and reliable control and guidance systems. These systems must effectively handle various encounters
with both manned and unmanned vessels, and operate under diverse weather and sea conditions. A
significant challenge in this endeavor is ensuring that autonomous vessels comply with the International
Regulations for Preventing Collisions at Sea (COLREGs).
Following the creation of a new, simple yet realistic Python simulator last year, the aim of this work
was to further develop the simulator, focusing particularly on the implementation of different path
planning algorithms. In addition to the initially present Artificial Potential Field method, four other
algorithms were implemented: A* (A Star), D* Lite (for dynamic), Ant Colony Optimization, and
Particle Swarm Optimization. The second part of the work involved comparing these algorithms and
evaluating their efficiency in different situations. Criteria like the length of the path and the runtime of
the algorithms were measured in fixed environments. Various two-boat encounters were also examined
to test how well the algorithms can quickly replan in dynamic environments and avoid collisions at sea.
Arthur Coron
Dates : Du 30/04/2024 au 30/07/2024
Laboratoire ou Entreprise : University of Oslo
Confidentiel : Oui/Non
Lieu : Forskningsveien 3A, Sondergaard, Norvege
Titre du stage : prototyping and control software development for our robot platforms
Encadrant : Pia Sondergaard
Résumé : ???
Celian Daligault
Dates : Du 13/05/2024 au 01/09/24
Laboratoire ou Entreprise : Rutgers University
Confidentiel : Oui/Non
Lieu : Rutgers, The State University of New Jersey 57 US Highway 1, New brunswick, Etats-unis
Titre du stage : Design and implementation of advanced control methods on aerial robots
Encadrant : Laurent Burlion
Résumé : ???
Nicolas Damageux
Dates : Du 29/04/2024 au 16/08/2024
Entreprise : Airbus operations SAS Toulouse
Confidentiel : Oui
Lieu : Site de Saint Martin 316, route de Bayonne, Toulouse, France
Titre du stage : Modélisation et optimisation des performances avion au décollage (algorithme du simplex, réseaux de neurones)
Encadrant : Serge Laporte
Résumé : L'entraînement des modèles de Machine Learning reste un enjeu crucial lors du développement d'outils se basant sur des réseaux de neurones.
En particulier, les modèles utilisés pour la régression ont souvent besoin de jeux d'apprentissage complets et réellement descriptifs des variations des fonctions considérées.
De plus, il est vital de manipuler des modèles dont la complexité et l'architecture restent cohérents avec le niveau de précision nécessaire.
Dans ce contexte, le département d'étude des performances au décollage d'Airbus Commercial Aircraft (Toulouse) s'est rapidement rendu compte de la simplicité des modèles utilisés
pour le calcul de nombreuses performances. En particulier, les architectures et méthodes implémentées dans les années 19900 n'ont pas réussi à s'adapter à l'évolution des avions
durant les décennies qui suivirent, ce qui amène le département à manipuler des outils peu performants et qui commettent parfois des erreurs.
Cette étude a pour principal objectif de mettre en avant diverses approches permettant l'amélioration desdits modèles de Machine Learning.
En particulier, la notion d'entraînement sera abordée, et la représentation du domaine de décollage des avions sera au centre de ce rapport.
Nous verrons qu'il existe plusieurs méthodes permettant d'identifier des ensembles mathématiques qui ne sont pas connus explicitement,
ainsi que de les manipuler pour générer des bases de données complètes.
Marie Dubromel
Dates : Du 01/05/24 au 25/08/2024
Laboratoire ou Entreprise : CROSSING IRL
Confidentiel : Non
Lieu : Australie
Titre du stage : COLREGs Simulator
Encadrant : Benoît Clément
Résumé : The rising popularity of autonomous vessels necessitates the development of robust collision avoidance systems
to ensure safe navigation in unpredictable maritime environments. This internship report explains the design and
implementation of two simulators, each aimed at addressing the challenges associated with collision avoidance
while adhering to International Regulations for Preventing Collisions at Sea (COLREGs). The first simulator,
Simulator A, employs artificial potential fields to identify potential collision scenarios, calculate appropriate vector
fields, with an adapted controllers. It offers a comprehensive approach to collision avoidance, incorporating real-
time monitoring systems to ensure adherence to COLREGs during simulated scenarios. By allowing users to select
from a range of agent types representing different categories of unmanned surface vessels (USVs), it facilitates
analysis by experienced sailors. In contrast, Simulator B utilises real Automatic Identification System (AIS) data
from past USV or vessels trips, enabling the seamless integration of fictive and real-world USV scenarios. By
leveraging machine learning techniques, particularly deep reinforcement learning (DRL), these simulators pave
the way for adaptive and model-independent collision avoidance systems in various environmental conditions. In
conclusion, the development of these simulators help to progress towards building collision avoidance systems
that comply with COLREGs for manned, unmanned, and hybrid vessels. By fostering collaboration between
private companies and research laboratories, these advancements hold the promise of unlocking new possibilities
for safe and efficient maritime exploration. Moving forward, continued refinement and integration of AI-driven
enhancements will be crucial in ensuring the safety and efficacy of autonomous and hybrid vessel operations,
ushering in a new era of maritime innovation and exploration.
Jury : Luc Jaulin, Andreas Rauh, Marit Lahme
Gabriel Dugué
Dates : Du 06/05/24 au 19/07/2024
Laboratoire ou Entreprise : International University - Ho Chi Minh City
Confidentiel : Oui/Non
Lieu : , Ho chi minh city, Vietnam
Titre du stage : Stage à l'université internationale d'éléctronique d'Ho Chi Minh Ville
Encadrant : Huynh Tan Quoc
Résumé : ???
Thibault Edouard
Dates : Du 06/05/24 au 23/08/2024
Laboratoire ou Entreprise : Envico Technologies Limited
Confidentiel : Non
Lieu : De Havilland Way, Mount Maunganui, Tauranga, New-Zeland
Titre du stage : Design of a remote device to autonomously detect rats in Guatemala sugar can fields
Encadrant : Cameron Baker
Résumé : This report presents the development of an automated rodent detection system incorporating a multi-sensor approach
to accurately identify rats while minimizing false positives from other animals.
The system integrates two Infrared (IR) Sensor Modules, a temperature sensor, and an ESP32-CAM
for visual verification. The IR Sensor Module, comprising an IR emitter and
receiver, serves as the primary detection mechanism. To enhance accuracy,
two IR modules are used in tandem to confirm the presence of an object, while
the MLX90614 temperature sensor distinguishes between warm-blooded rats
and cold-blooded reptiles, such as snakes. The data from these sensors are
processed by an Arduino Nano, which sends a signal to the ESP32-CAM.
The ESP32-CAM captures and stores images for manual or AI-based verification during the trial phase. The choice of these sensors is justified by
their affordability and sufficient accuracy for initial testing. Additionally, a
Real-Time Clock (DS3231) is employed to timestamp detections, support-
ing integration with a database for comprehensive data management. This
multi-faceted approach aims to create a robust detection system capable of
effectively distinguishing rats from other animals, with considerations for cost
and accuracy. Furthermore, a graphical user interface (GUI) was designed
using Qt Designer with Python, allowing for user-friendly interaction and
visualization of detection data. The final system leverages these components
to achieve robust rodent detection, communication, and data management,
aligning with the project’s goals.
Juliette Faury
Dates : Du 13/05/2024 au 01/09/24
Laboratoire : Intelligent Robotics Lab - UVA
Confidentiel : Non
Lieu : University of Amsterdam Science Park 900, Amsterdam, Pays-bas
Titre du stage : Implementation of SLAM on the KUKA youBot robot of the Intelligent Robotics Lab of Amsterdam
Encadrant : Arnoud Visser
Résumé : The KUKA youBot is a mobile manipulator that was primarily developed for educa-
tion and research. It’s consists of two main parts:
— The KUKA youBot omni-directional mobile platform consists of the robot chassis,
four mecanum wheels, motors, power and an onboard PC board. Users can either
run programs on this board, or control it from a remote computer.
— The KUKA youBot arm has five degrees of freedoms (DOF) and a two-finger
gripper. If connected to the mobile platform, the arm can be controlled by the
onboard PC. Alternatively, the arm can be controlled without the mobile platform
by using an own PC connected via Ethernet cable.
Additional sensors can be mounted on the robot. The purpose of my internship was
to implement a SLAM (Simultaneous location and mapping) method with a Velodyne
Lidar. During the first weeks, ROS Hydro was accidentally removed from the robot.
The rest of the internship consisted in trying to reinstall correctly ROS (the sofware
wasn’t available anymore in the usual download methods, because it had become too
outdated). I suceeded in partially restoring the software, but some parts of it were still
not functioning properly.
Jury : Marit Lahme, Luc Jaulin
Christian-Joël Gbikpi
Dates : Du 06/05/24 au 06/09/24
Laboratoire ou Entreprise : Colorado State University - Cal State - CSU
Confidentiel : Oui/Non
Lieu : Composite Material, Manufacture and Structure Laboratory 3317, Fort collins, Etats-unis
Titre du stage : Robot de chirurgie musculo-squelettique
Encadrant : Kirk Mcgilvray
Résumé : During my internship at OBRL I had the opportunity of working on 2 distinct
mission on the Universal Robot 5.
First I had to program the robot so that it could point an axis in space to
a surgeon during an ACL replacement procedure. This implied getting the
geometrical information and commanding the displacement in a way that was
safe to the operators.
Second I had to try to use the arm to get a limb through a path while
exerting a force. Notwithstanding my failure to get a functioning result, I
experimented through several ways to accomplish the task and suggested way to
make my first draft work
Arne Jacobs
Dates : Du 06/05/24 au 26/07/2024
Laboratoire ou Entreprise : Volvo Car Corporation Headquarters
Confidentiel : Oui/Non
Lieu : Gunnar Engellaus väg 8, Göteborg, Suède
Titre du stage : Ensure smooth transition from simulation to Fanuc robot controller downloads
Encadrant : Fredrik Nagard
Résumé : ???
Laura Jouvet
Dates : Du 06/05/24 au 23/08/2024
Laboratoire : Hambourg University
Confidentiel : Non
Lieu : Mittelweg 177, Hamburg, Allemagne
Titre du stage : Integration of monocular depth estimation and implementation of open vocabulary visual object detection in NICO and NICOL robots
Encadrant : Cornelius Weber
Résumé : The aim of this internship was to find a research topic in the field of object detection with
NICOL robot and to work on this subject. The internship involved algorithms development
and testing in a simulated environment using ROS and Gazebo, integration of 3D cameras on
NICOL robot and research about objects detectors.
This multidisciplinary internship encompassed simulation, hardware integration, project
management and collaboration within a research group.
Jury : Marit Lahme, Luc Jaulin
Louise Lapie
Dates : Du 01/05/24 au 21/08/2024
Laboratoire ou Entreprise : AMOLF
Confidentiel : Oui/Non
Lieu : Université d'Amsterdam, Science Park 104, Amsterdam, Pays-bas
Titre du stage : Conception d'un bioreacteur robotisé pour un coeur artificiel
Encadrant : Bas Overvelde
Résumé : ???
Titouan Leost
Dates : Du 06/05/24 au 23/08/2024
Laboratoire : Aston University
Confidentiel : Non
Lieu : Aston st, Birmingham, Royaume-uni
Titre du stage : Development of a software architecture for an autonomous sailboat
Encadrant : Jian Wan
Résumé : The aim of the project was to develop an easily reusable software
architecture to enable a sailing boat to carry out autonomous missions. In addition, the idea was to
use the simplest possible hardware to provide a low-cost, low-tech solution.
The project therefore initially involved setting up an architecture to collect and process the data
from the various sensors in order to feed an algorithm controlling the sailboat via servomotors.
After ensuring the sensors were functioning correctly, the robustness of the architecture was verified
by conducting tests on a lake, using a simple algorithm whose effectiveness had already been proven.
Jury : Andreas Rauh and Luc Jaulin
Simon Martineau
Dates : Du 06/05/24 au 23/08/2024
Laboratoire: Tallinn University of Technology
Confidentiel: Non
Lieu : Ehitajate tee 5, Tallinn, Estonie
Titre du stage : Navigation et localisation d'un robot avec differentes techniques SLAM.
Encadrant : Roza Gkliva
Résumé : This report documents the work I performed over 4 months between May
and August 2024 as part of my ERASMUS internship. I participated
in the ROBOMINERS project at the Center for Biorobotics at TalTech
under the supervision of Roza Gkliva. The goal of this internship was to
develop a program to evaluate the quality of maps being created by SLAM
algorithms aboard the RM3 robot during underground exploration. The
first part of this report shows my initial discovery of SLAM algorithms
and how I set up the hardware and software for this project. The second
part goes over the different algorithms I developed to simulate noise and
virtual dust clouds that could affect the robot’s sensors. The final part
goes over the experimentation phase and the conclusions I drew from the
different tests I performed.
Basile Mollard
Dates : Du 06/05/24 au 25/08/2024
Laboratoire : The Intelligent Robotics Laboratory, University of Amsterdam
Confidentiel : Non
Lieu : Science Park 900, Amsterdam, Pays-bas
Titre du stage : SLAM with the Velodyne Puck Lidar on the KUKA youBot
Encadrant : Arnoud Visser
Résumé : As part of my second year of engineering studies in robotics at ENSTA Bretagne, I completed a
three-month internship at the University of Amsterdam. The project aimed to develop autonomous
navigation for the Kuka Youbot robot, in collaboration with Juliette Faury. Our main objective
was to enable the robot to navigate autonomously using a Velodyne VLP16 Lidar combined with
a camera, applying Simultaneous Localization and Mapping (SLAM) techniques. This work built
on previous research that used only a camera for SLAM, but our approach was to integrate Lidar
data to enhance accuracy.
I was responsible for calibrating the Lidar and developing a ROS node to acquire and process
the data, while Juliette focused on integrating the camera and setting up the robot. Unfortunately,
due to an error that occurred after the first month of work, we were unable to fully achieve our
initial objectives, despite three months of attempts to resolve the issue.
Jury : Marit Lahme, Luc Jaulin
Antoine Morvan
Dates : Du 01/05/24 au 31/08/2024
Laboratoire : Tokyo University of Science (TUS)
Confidentiel : Non
Lieu : Niijukuku, Katsushika-ku, Tokyo, Japon
Titre du stage : Development of a low-cost acoustic underwater communication system
Encadrant : Takuya Hashimoto
Résumé : As part of my second year of engineering school, I completed an approximately 18-week in-
ternship at the Tokyo University of Sciences on the Katsushika-ku campus in the Hashimo-
toLab robotics laboratory. During this internship I worked on the development of a low
cost underwater acoustic communication system to be able to equip eventually small un-
derwater robot.
The difficulty of communicating with robots in an underwater environment where
most other conventional means of communication (Wi-Fi, GPS, radio,.) do not work
is a well known problem. Today, one of the preferred solutions in this field is acoustic
communication by ultrasound, however, if there are already systems for this, they are for
the most part very expensive and very often rely on patented technologies making their
operation and adaptation often complicated. This project aims to respond precisely to
this problem.
Jury : Andreas Rauh, Luc Jaulin
Océan Noël
Dates: Du 01/11/23 au 31/08/2024
Laboratoire: CNRS-AIST JRL
Confidentiel : Non
Lieu : AIST Tsukuba Central 1, Umezono, Tsukuba, Ibaraki, 305-8560, Japan
Titre du stage : Haptics Technology Enhancement for Avatar Robot Teleoperation
Encadrant : Rafael Cisneros-Limón
Résumé : I present my work at CNRS-AIST JRL, where I served both as an Intern
and a Research Assistant (RA). During this internship, I had the opportunity to work on two
distinct subjects. The first, as an intern, involved integrating the complete software for an
omnidirectional robot equipped with a mounted arm, designed to collect cardboards and
trash bags in shopping malls. This work encompassed low-level algorithms for sensor
data acquisition and actuator control, as well as high-level perception, mapping, and
navigation features. My main contribution is a novel approach for efficient 2D navigation
using a multimodal sensor fusion technique, implemented as a ROS2 package. This work
resulted in a paper submission to the robotic conference SII 2025. As a Researcher
Assistant, my project centered on conducting research and development related to
Avatar robot technology, aimed at transporting human presence and senses to a remote
location in real time. A critical aspect of this technology is the implementation of a proper
haptics system, enabling the operator to feel pressure, texture, and temperature at a
remote location through the avatar robot. My research in this area led to the development
of a novel concept for an omnidirectional robot on a soft spheric roller and the creation of
a Coaxial Multi-Channel Rotary Transmission System. We are currently in the process of
applying for a patent with CNRS-AIST JRL for this system.
Jury : Luc Jaulin
Camilo Ortiz
Dates : Du 06/05/24 au 30/08/2024
Laboratoire : INDESS, University of Cadiz
Confidentiel : Non
Lieu : Cadiz, Espagne
Titre du stage : Interval analysis and fuzzy numbers
Encadrant : Manuel Arana-Jiménez
Résumé : I will talk about my role during the internship, how little by little my main objectives became clearer for me,
how I dealt with some issues and how I approached my different
tasks. The internship, as challenging as it was, came with its share of issues and questions.
We will see in this report how I think it will help in my growth as an engineer. We will
also talk about interval analysis and fuzzy numbers, as these were the main subjects of my
internship. I will provide a few snippets of my code. Finally, we will talk about research
and plagiarism, which is one of the most import subject that raised my attention during the
internship.
Jury: Luc Jaulin
Gaétan Pérez
Dates : Du 06/05/24 au 30/08/2024
Laboratoire : NORCE
Confidentiel : Non
Lieu : Grimstad, Norvège
Titre du stage : Robotics for Automated Inspection using a Multimodal Sensor Gimbal
Encadrant : Atle Aalerud
Résumé : Railway inspection is essential to ensuring the safe and efficient operation of rail networks. Regular
inspections help detect structural faults around railway lines, preventing accidents that could lead
to severe damage. As part of the European IAM4RAILS project, a collaboration between NORCE
and SNCF is developing a robot to automate this process. The robot uses various sensors, including
cameras and LIDARs, to inspect cables, electric pillars, rails, and tunnels. Innovative techniques,
such as cable reflection detection to estimate wear, and 2D LIDAR for rail and cable positioning, are
being explored. A tunnel scanning system is also being developed to identify structural issues. To
validate these solutions, a field test was conducted, recording data from the sensors. The control of
these sensors by a user is an integral part of the project, with different options explored such as using
a basic gamepad, or a SteamDeck.
Jury : Andreas Rauh and Luc Jaulin
Simon Philibert
Dates : Du 06/05/24 au 26/07/2024
Laboratoire ou Entreprise : Institute for Research and Robotics (ISR LISBOA)
Lieu : Dynamical Systems and Ocean Robotics Lab (DSOR) ISR – Instituto Superior Técnico, Torre Norte – 7º Piso Av.Rovisco Pais, Lisboa, Portugal
Titre du stage : Slocum underwater glider navigation
Encadrant : David Cabecinhas
Résumé :
The key objectives are to:
1. Learn the methodologies used to develop a kinematic / kinematic+dynamic model of an underwater glider subjected to currents. Idem, regarding the
methodologies used for system identification.
2. Model the SLOCUM glider.
3. Learn selected methods of vehicle control and apply them to the design of a controller for the SLOCUM glider using its backseat driver interface
4. Become familiar with the software tools developed at ISR-IST for marine vehicle systems simulation and visualization using Matlab,
C++, the ROS-1 operating systems, and Gazebo. Assess the performance of model identification and control systems in simulation.
5. Plan and execute in water experiments for glider model identification and control
6. Write a full report on the work done
Knowledge, skills and competences to be acquired by the end of the traineeship (expected Learning Outcomes):
- In depth understanding of the methodologies for underwater glider navigation and control
- Become familiar with tools available for system hardware-in-the-loop simulation and performance assessment
- Acquire experience in the realization of field trials
Ylona Provot
Dates : Du 01/05/24 au 31/08/2024
Laboratoire ou Entreprise : ZEISS Innovation HUB
Confidentiel : Oui
Lieu : Hermann-von-Helmholtz-Platz 6, Eggenstein-leopoldshafen, Allemagne
Titre du stage : Medical Robotics with focus on experiments
Encadrant : Ernar Amanov
Résumé : The aim is to determine if a robot originally designed for retina surgery could be used in the anterior segment of the eye to
perform Minimally Invasive Glaucoma Surgeries.
The work consists in determining the workspace covered by the robot and see if the whole anterior segment
could be reached by the end-effector.
Jury : Marit Lahme and Luc Jaulin
Harendra Rangaradjou
Titre du stage : Développement d'un voilier autonome
Confidentiel : Non
Laboratoire : Aston University
Lieu : Aston St, Birmingham B4 7ET, UK
Encadrant : Jian Wan
Tuteur école : Luc Jaulin
Résumé : Je contribuerai au développement et à l'implémentation de nouveaux algorithmes de correction et/ou d'apprentissage pour un voilier miniature
autonome en utilisant des capteurs, microcontrôleurs et logiciels de pointe
Jury : Taddéo Guérin, Luc Jaulin
Aimé Randriamoramanana
Titre du stage : Modélisation et Simulation d’un Véhicule Sous-Marin Autonome
Confidentiel : Oui
Laboratoire : University of Adelaide
Lieu : Adelaide, Australie
Encadrant : Eric Fusil
Tuteur école : Luc Jaulin
Résumé : Ce rapport présente le développement et la simulation d’un véhicule sous-marin autonome (AUV)
nommé Harmonia II, dans le cadre du projet ARES à l’Université d’Adélaïde. Le travail a débuté par
une analyse approfondie des différents simulateurs disponibles, afin de choisir celui le plus adapté aux
besoins spécifiques de l’expérimentation en laboratoire. L’objectif principal était de sélectionner un
simulateur capable de reproduire fidèlement les conditions réelles auxquelles le sous-marin est confronté,
tout en permettant des ajustements pour les futures configurations expérimentales.
Une fois le simulateur sélectionné, l’attention s’est portée sur la modélisation et le contrôle du
sous-marin dans des scénarios simulés plus complexes. Cela inclut la réflexion sur l’intégration de
systèmes de commande avancés, permettant d’automatiser et de gérer des missions complexes sous
l’eau, tout en prenant en compte les contraintes physiques et dynamiques du véhicule. La simulation
joue ainsi un rôle clé dans la validation des algorithmes de contrôle avant leur déploiement sur le
modèle physique en laboratoire, car elle permet également de déterminer certaines problématiques liées
au positionnement exact du robot.
Jury : Luc Jaulin, Marit Lahme
Ambre Ricouard
Dates : Du 06/05/24 au 23/08/2024
Laboratoire ou Entreprise : TU WIEN
Confidentiel : Non
Lieu : Wien, Autriche
Titre du stage : Combination of Stochastic Model Predictive Control and Vision System on a Donkey Car for Circuit Following Task
Encadrant : Minh-Nhat Vu
Résumé : This project focuses on automating a small-scale model car to navigate a fixed circuit using
only a camera for guidance. The objective is for the car to follow a track centered between two
guiding lines. The circuit used in reality is also available in simulation, allowing for process
validation before testing in a real environment where significant visual noise may be present.
A crucial initial step is to examine the system specifications and camera limitations to
anticipate all possible scenarios, thereby ensuring that the developed controllers are robust.
Depending on the car’s position and orientation relative to the track, it may detect both lines,
only one line, or even none. Therefore, the system must be able to handle each of these situations
effectively.
The process begins with image processing, where the visible lines are detected to determine
the car’s orientation and desired future position. With this information in hand, the next step is
to address vehicle control. This requires establishing the car’s kinematic model, which involves
defining a projection frame and understanding how the system inputs relate to the vehicle’s
kinematics.
Initially, a simple Proportional-Integral-Derivative (PID) controller is implemented. While
this provides a good starting point, it is inherently limited as it only accounts for past positions. To
improve system robustness, a stochastic Model Predictive Control (MPC) algorithm is introduced,
enabling for the prediction of future trajectories and more effective decision-making.
Jury : Andreas Rauh, Luc Jaulin
Léa Rion
Dates : Du 06/05/24 au 02/08/24
Laboratoire ou Entreprise : Bioengineering Laboratory of the CEU, University of Madrid
Confidentiel : Oui/Non
Lieu : Madrid, Espagne
Titre du stage : Rehabilitation Robotics
Encadrant : Raya López Rafael
Résumé : ???
Salah El Din Sekar
Dates : Du 02/05/24 au 30/08/2024
Laboratoire ou Entreprise : FLORALIS - UGA Filiale
Confidentiel : Non
Lieu : 7 Allée de Palestine, Gieres, France
Titre du stage : Imitation Control Learning for Robotic Systems using Neural Networks
Encadrant : Thao Dang
Résumé : This report represents my internship at Verimag, Grenoble, under the supervision of Dr. Thao Dang.
The project focuses on using imitation learning to train a neural network controller that integrates the
advantageous properties of conventional controllers specified using Signal Temporal Logic (STL). The
implementation is applied to robotic systems utilizing the Robot Operating System (ROS2).
The core of the project involves creating high-quality data for training a neural network (NN) controller
by generating behavioral traces through ROS2 simulations. By high-quality data, we mean data that
is both diverse and comprehensive, ensuring good generalization ability for the resulting neural network
controller. These traces are then used to train the neural network using MATLAB. To ensure the accuracy
and reliability of the NN controller, the Breach tool is employed for falsification, identifying and rectifying
deviations from expected behaviors. This approach aims to enhance the accuracy, robustness, and safety
of the neural network controllers, making them suitable for deployment in complex, real-world robotic
applications.
Jury : Luc Jaulin, Marit Lahme
Matti Soucaille
Dates : Du 29/04/2024 au 27/09/2024
Laboratoire ou Entreprise : SPACEDREAMS
Confidentiel : Oui/Non
Lieu : 20 rue Ampère, Massy, France
Titre du stage : Stage maquette d’un système de contrôle commande
Encadrant : Raphaël Methot
Résumé : ???
Luc-André Terrine
Dates : Du 29/04/2024 au 27/09/2024
Laboratoire : University of Oldenburg
Confidentiel : Non
Lieu : Carl Von Ossietzky University, Oldenburg, Germany
Titre du stage : Modeling the robot ViperX 300
Encadrant : Andreas Rauh
Résumé : As part of my formation at ENSTA Bretagne, I was required to complete an internship
in a foreign country as an engineering assistant. I was given the opportunity to fulfill this
requirement at the University of Oldenburg, where I joined their research team. My work
focused on optimizing the code for computing the equation of motion of a robotic arm, with
the goal of enabling the forward dynamics simulation of its multiple degrees of freedom.
Jury : Luc Jaulin, Andreas Rauh, Marit Lahme
Main Tihami Ouazzani
Dates : Du 14/05/2024 au 04/09/24
Laboratoire ou Entreprise : Colorado State University - Cal State - CSU
Confidentiel : Non
Lieu : Composite Material, Manufacture and Structure Laboratory 3317, Fort collins, USA
Titre du stage : Real-Time Robotized Spine fusion Testing with UR5e System
Encadrant : Kirk McGilvray
Résumé : This project focuses on the integration and control of a UR5e robotic system for conducting
orthopedic research, specifically through testing bone samples. The primary goal is to
simulate pseudo loads on bone samples using real-time data to assess mechanical responses
under various conditions. The project utilizes the AMTI MC3A transducer and Gen5
amplifier, alongside the UR5e robot, for precise data collection and control.
Jury : Andreas Rauh and Luc Jaulin
Adrian Vanalli Carraro
Dates : Du 06/05/24 au 23/08/2024
Laboratoire ou Entreprise : STIIMA - CNR
Confidentiel : Non
Lieu : via G. Amendola 122/D, Bari, Italie
Titre du stage : Robot 4x4 d'utilisations générales
Encadrant : Annalisa Milella
Résumé : This report presents the development and implementation of a vine-following
system for an autonomous exploration robot, part of a research project at the mobile
robotics laboratory of the CNR in Italy. The main objective was to migrate the robot’s
software to ROS2 to enhance performance, modularity, and the future integration of
more complex algorithms, such as AI-based ones. The robot, equipped with advanced
sensors like the Intel RealSense camera and high-precision GPS, uses a chain of
nodes to analyze its 3D environment and adjust its trajectory. The work also includes
collaboration with another robot, the Pioneer, to test "swarm" functionalities in a
precision agriculture context.
Jury : Luc Jaulin