|
|
|
|
|
|
|
|
|
|
Alix Agnes
Title : Comparison between interactive robots and screens in social assistance
Reference paper :
B.J. Zhang, R. Quick, A. Helmi, N.T. Fitter.
"Socially Assistive Robots at Work: Making Break-Taking Interventions More Pleasant, Enjoyable, and Engaging."
In: 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)
Abstract :
The goal of socially assistive robots is to provide assistance to humans in need of their services,
and this assistance is carried out via social interaction. These services can be used in many settings : education, healthcare, entertainment...
But as their use becomes more and more widespread, the question of their improved usefulness as opposed to screens or tablets accomplishing
the same actions is an important one. This article intents to compare the benefits and drawbacks of both robots and screens in social settings.
There are many factors that come into play when it comes to this comparison, such as the cost of production,
its ease of use, the perception users have of these interfaces, its flexibility...
The end goal of this article is to determine which of these two types of human-machine interfaces (HMI) is more relevant for users.
Simon Barbarit-Gaboriau
Title : Real-Time Metric-Semantic Localization and Mapping: Exploring the Integration of Semantics in Visual SLAM
Reference papers : https://arxiv.org/abs/1910.02490 et https://arxiv.org/abs/2209.06428
Abstract :
This paper explores the transformative potential of incorporating semantic information into traditional visual simultaneous localization and mapping (SLAM) algorithms.
By integrating semantic segmentation networks into the SLAM pipeline, our aim is to enable the identification and labelling of objects and scene elements
in real time, enhancing robot perception and decision-making capabilities especially in dynamic and complex environments.
Through a detailed analysis, we illustrate how adding semantic layers to purely geometric maps improves environmental understanding, facilitates better interaction
with dynamic and unstructured scenes, and contributes to more efficient navigation and planning.
The paper discusses the framework, implementation challenges, and potential applications of this approach, offering insights into the future of intelligent mapping and localization systems.
Romain Bornier
Title : Combinaison de l'Apprentissage Interactif et du Deep Reinforcement Learning pour la Planification de Trajectoire des AUVs
Reference paper : Action Guidance-Based Deep Interactive Reinforcement Learning for AUV Path Planning, Dong Jiang, Zheng Fang, Chunxi Cheng, Bo He, Guangliang Li, 2022 International Conference on Machine Learning, Control, and Robotics (MLCR)
Abstract :
Une nouvelle méthode de planification de trajectoire pour véhicules sous-marins autonomes (AUV) est proposée, combinant l'apprentissage interactif avec
l'algorithme Deep Deterministic Policy Gradient (DDPG) pour créer l'algorithme Interactive DDPG (IDDPG). Contrairement aux approches classiques,
IDDPG intègre les suggestions humaines afin d'orienter les actions des AUV et d'ajuster les récompenses en temps réel, ce qui permet d'améliorer l'efficacité de l'échantillonnage.
Les tests effectués sur la plateforme de simulation Gazebo montrent que cette méthode offre une meilleure généralisation et un apprentissage accéléré par rapport aux méthodes traditionnelles.
Ce développement ouvre des perspectives pour des systèmes plus autonomes et robustes dans des environnements marins complexes.
Arthur Coron
Title : Deep blue AI: A new bridge from data to knowledge for the ocean science
Reference paper : https://www.sciencedirect.com/science/article/pii/S0967063722001984
Abstract :
The article examines how artificial intelligence (AI) can bridge oceanic "big data" (blue data) and knowledge for sustainable ocean management.
It reviews AI frameworks like feature engineering, detection, and time-series forecasting, with examples such as Arctic ice prediction and marine debris detection.
Challenges include the need for labeled data, explainable AI models, and interdisciplinary collaboration.
The study emphasizes that leveraging AI can unlock valuable insights from complex oceanic data, advancing both ocean science and AI technologies.
Marie Dubromel
Title : Comparative Evaluation of Path Planning Algorithms for Autonomous Vessels in Dynamic Maritime Environments
Reference paper : Benoit Clement, Marie Dubromel, Paulo E. Santos, Karl Sammut, Michelle Oppert, and Feras Dayoub. Hybrid navigation acceptability and safety. Proceedings of the AAAI Symposium Series, 2(1):11–17, 2023.
Abstract :
This paper presents a comprehensive comparison of path planning algorithms for autonomous vessels, focusing on their efficiency and adaptability in dynamic maritime environments.
Building on a previously developed Python-based simulator using the Artificial Potential Field (APF) method, this work extends the simulator
by integrating four additional path planning algorithms: A* (A Star), D* Lite (for dynamic environments), Ant Colony Optimization, and Particle Swarm Optimization.
The primary objective is to evaluate these algorithms based on their ability to plan efficient, collision-free paths while navigating complex maritime scenarios, including encounters with both manned and unmanned vessels. The performance of each algorithm is assessed through multiple criteria, including path length, computational runtime, and their ability to replan in real-time during dynamic vessel encounters. This paper provides a detailed analysis of how each algorithm performs in fixed and dynamic environments, offering valuable insights for enhancing the autonomy and safety of maritime navigation in compliance with the International Regulations for Preventing Collisions at Sea (COLREGs).
Juliette Faury
Title : Control Strategy for the Autonomous Navigation of a Ducted Fan Flying Robot
Reference paper : J. M. Pflimlin, T. Hamel, P. Soueres and R. Mahony,
"A hierarchical control strategy for the autonomous navigation of a ducted fan flying robot"
Proceedings 2006 IEEE International Conference on Robotics and Automation, 2006. ICRA 2006., Orlando.
Abstract :
Ducted Fan are small and discreet secure platforms, which are able to perform vertical takeoff and landing (VTOL) and stationary flight.
A ducted-fan UAV is a UAV that has similar rotors configuration with a coaxial helicopter, but the rotors are mounted within a cylindrical duct.
The duct helps to reduce thrust losses of the propellers, and the ducted fans normally have rotational speed.
They are of evident interest for civil and military operations in an urban environment
(shrouding a rotating propeller protects against blade strikes with objects at low altitude and people around the aircraft).
However, this kind of vehicle is unstable and its dynamics along the three axes are strongly coupled.
The purpose of this paper is to simulate and compare several control strategies for the autonomous navigation of a ducted fan UAV
(based on the HoverEy ducted fan), with for instance hierarchical controller or Lyapunov control functions.
Laura Jouvet
Title : Adapting Segment Anything Model for water column images of multibeam sounder
Reference paper : Lin Wang, Xiufen Ye, Liqiang Zhu, Weijie Wu, Jianguo Zhang, Huiming Xing, and ChaoHu. "When sam meets sonar images", 2023.
Abstract :
Image segmentation has become indispensable in many fields of computer vision like autonomous vehicles, security or augmented reality. Since April 2023,
the Segment Anything Model (SAM) created by Meta has revolutionized the way of segmentation. Its two main highlights are its impressive capabilities in various segmentation tasks and its prompt-based interface.
However, although SAM is really efficient with natural images, it has a hard time segmenting other types of images. Some researchers succeeded in obtaining promising results for medical images
by employing fine-tuning techniques but there is still a lack of research with sonar and multibeam sounder images. This paper provides a comprehensive investigation of SAM's performance
on water column images of multibeam sounder.
First we evaluate SAM using various settings on water column images and then we fine-tune SAM in order to obtain significant improvement in the performance of fine-tuned SAM with water column images.
Titouan Leost
Title : A comprehensive review of unmanned ground vehicle terrain traversability in unstructured environments
Reference paper : https://www.sciencedirect.com/science/article/pii/S2215098623001350?ref=pdf_download&fr=RR-2&rr=8e8ad9f6d93c02b1
Abstract :
This article presents a comprehensive analysis of unmanned ground vehicle (UGV) terrain traversability assessment, structured into three main areas:
terrain classification, terrain mapping, and cost-based traversability.
These are further divided into appearance-based, geometry-based, and hybrid methods.
The study also takes a look at the role of machine learning (ML), deep learning (DL), reinforcement learning (RL),
and other end-to-end approaches, as pivotal tools for enhancing terrain traversability analysis.
The findings highlight the advantages of combining exteroceptive and proprioceptive sensors for a more efficient, optimized, and reliable assessment.
Overall, this paper aims to make a significant contribution to advancing the understanding of traversability analysis in unstructured environments.
Basile Mollard
Title : Titre
Reference paper : ???.
Abstract :
Antoine Morvan
Title : Explicit vs. Tacit Leadership in Influencing the Behavior of Swarms
Reference paper : ???.
Abstract :
De nombreux chercheurs utilisent un leader téléopéré pour influencer un essaim robotique, mais les mécanismes précis de cette influence restent peu étudiés.
Certains adoptent des leaders désignés, clairement identifiés comme tels par les autres membres de l’essaim et suivis.
D’autres n’imposent pas de distinction leader/suiveur et influencent l’essaim indirectement en contrôlant un ou plusieurs membres.
La robustesse des essaims reposant sur des interactions distribuées, un leader désigné peut les rendre sensibles au bruit ou perturber leur cohérence.
À l’inverse, un contrôle indirect par effets locaux peut être trop lent pour un contrôle humain efficace.
Cette étude compare deux méthodes : "flooding", où l’influence du leader prédomine, et "consensus", où les robots suivent la moyenne des voisins.
Les simulations montrent que le flooding converge plus rapidement mais est plus sensible au bruit, tandis que le consensus converge plus lentement mais améliore la connectivité.
Les tests avec des opérateurs humains confirment l’efficacité du flooding pour atteindre des objectifs, bien que l’erreur dégrade de manière similaire les deux méthodes.
Cependant, le consensus démontre des avantages clairs en termes de cohésion et de connectivité globale de l’essaim, notamment dans des conditions de capteurs imparfaits.
Océan Noël
Title : A Lightweight Approach to Efficient Multimodal 2D Navigation and Mapping: Unified Laser-Scans as an Alternative to 3D Methods
Reference paper : https://www.researchgate.net/publication/4122019_Colored_2D_maps_for_robot_navigation_with_3D_sensor_data
Abstract : In this paper, we propose a novel approach for efficient 2D navigation using a multimodal sensor fusion technique.
Our method focuses on merging data from multiple sensors, such as LiDARs, cameras, and ultrasonic sensors, into a unified Laser-Scan,
which serves as a foundation for faster and more lightweight navigation. By fusing sensor data at the Laser-Scan level,
our approach enables the use of basic 2D Simultaneous Localization And Mapping (SLAM) algorithms for mapping tasks, or any others Laser-Scan based features,
while still benefiting from the rich information provided by multimodal 3D inputs. This results in a more computationally efficient solution
compared to traditional 3D methods that rely on depth points or full multimodal SLAM systems.
Our experimental results demonstrate that the proposed approach achieves comparable accuracy in mapping and localization while significantly
reducing computational complexity and processing time. This research offers a promising alternative for real-time 2D navigation
in resource-constrained autonomous systems, such as drones or any small unmanned vehicles.
Camilo Ortiz
Title : Swarm of robots
Reference paper : https://royalsocietypublishing.org/doi/full/10.1098/rstb.2020.0309
Abstract : During a mission in an unknown environment, a swarm of robots needs to have a high potential of adaptivity in order
to realize the mission in the most effective way possible.
Swarms are well known for their ability to achieve complex tasks in spite of each individual robot having few control options.
When coupled with simulations of social learning, we can show that a swarm of robots can, in average,
learn and optimize its controller faster than if it were to learn alone. In this paper we discuss the forms of social learning that exist in the literature
and try to present, through a simulation, the different learning rates between those forms of learning.
Gaetan Perez
Title : Numerical modelling of optical tweezers applied to mobile micro-robots automatization
Reference papers :
[1]. E. Gerena, S. Régnier and S. Haliyo, "High-Bandwidth 3-D Multitrap Actuation Technique for
6-DoF Real-Time Control of Optical Robots," in IEEE Robotics and Automation Letters, vol. 4, no.
2.
[2]. H. Xin, Y. Li, Y.-C. Liu, Y. Zhang, Y.-F. Xiao, B. Li, Optical Forces: From Fundamental to
Biological Applications. Adv. Mater. 2020.
Abstract : Optical tweezers are contactless tools that enable precise manipulation of micron-sized
particles. Discovered in 1987, this technology can be integrated into robotic systems for
enhanced performance. This project aims to develop a numerical simulation to model the
interactions between optical tweezers and optical receptors, facilitating the automated control
of microrobots. Optical tweezers operate using a Gaussian light beam focused on a
microparticle, where the beam exerts a force similar to a harmonic oscillator, depending on the
particle's equilibrium position within the beam. The precise and automated control of such
microrobots has great potential across various fields, especially in biology, where accurate
manipulation is crucial for the success of delicate operations.
Simon Philibert
Title : Optimisation of robotic fish control (or schools of robotic fish) using Computational Fluid Dynamics (CFD)
Reference paper : Multi-Objective Multidisciplinary Design Optimization of a Robotic Fish System (CHEN 2021)
Numerical Simulation and Analysis of Fish-Like Robots Swarm (LI 2019)
Abstract :
The development of robotic fish has drawn considerable attention due to their potential applications in underwater exploration, environmental monitoring, and biological studies.
Inspired by the efficiency and adaptability of biological fish, these robots aim to replicate swimming mechanisms to achieve high performance in speed,
energy efficiency, and maneuverability. However, designing and optimising robotic fish is a complex multidisciplinary problem that requires integrating hydrodynamics, propulsion,
and control systems. Computational Fluid Dynamics (CFD) plays a crucial role in evaluating and refining these designs by simulating the interaction between robotic bodies and fluid environments.
Recent advancements suggest the potential for hybrid approaches, combining CFD with artificial intelligence techniques such as neural networks, to accelerate optimization processes.
This study seeks to explore and refine methods for optimizing the performance of individual robotic fish and their coordinated behavior as schools, focusing on multi-objective design strategies and hydrodynamic efficiency.
Aime Randriamoramanana
Title : Comparison of path planning solutions for autonomous ground vehicle in unstructured environment
Reference paper : Wang, Nan, Xiang Li, Kanghua Zhang, Jixin Wang, and Dongxuan Xie. 2024. "A Survey on
Path Planning for Autonomous Ground Vehicles in Unstructured Environments", Machines 12, no. 1: 31. https://doi.org/10.3390/machines12010031
Abstract :
Path planning in unstructured environments poses significant challenges for
autonomous ground vehicles (AGVs), particularly in terrains affected by heavy rainfall.
Such conditions introduce dynamic and complex factors, including reduced traction, and
obscured navigation landmarks. This paper focuses on the development and evaluation
of path planning algorithms adapted to AGVs operating in rain-soaked and muddy
environments. A comparative study of existing deterministic and learning-based
approaches is conducted, with a focus on their applicability to agricultural robots
navigating in muddy soils. The study includes testing and benchmarking different
solutions to identify the most effective methods for ensuring safe and reliable navigation
in these challenging conditions. By addressing the unique challenges of rain-affected
terrains, this work advances the understanding of autonomous navigation for
agricultural applications.
Harendra Rangaradjou
Title : A Comparison of RRT-connect, RRT*-connect and informed RRT*-connect Path Planning Algorithms
Reference paper : I. Noreen, A. Khan, and Z. Habib, “A comparison of RRT, RRT* and RRT*-smart path planning algorithms,”
International Journal of Computer Science and Network Security (IJCSNS), vol. 16, no. 10, p. 20, 2016
Abstract :
Path planning is a critical component in the navigation of autonomous mobile robots, ensuring efficient, collision-free traversal in dynamic and complex environments.
Sampling based planning algorithms derived from the Rapidly-exploring Random Tree (RRT) algorithm have been extensively studied in recent years.
They are probabilistic complete algorithms and are particularly well-suited for solving high-dimensional complex problems.
This paper focuses on the RRT-connect algorithm and two of its most popular extensions, namely RRT*-connect and informed RRT*-connect.
Firstly, a succinct analytical overview of the three algorithms is provided.
The impact of key parameters on algorithm performance is then evaluated.
Lastly, a performance comparison based on optimality criteria such as path cost, run time, and the total number of nodes in the tree is conducted
through simulation-based experiments, offering insights into their relative strengths and trade-offs.
Ambre Ricouard
Title : Adaptive Reinforcement Learning-Driven MPC for Efficient Autonomous Vehicle Navigation
Reference paper : https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9655042
Abstract :
This paper introduces an innovative approach to autonomous vehicle navigation by combining Deep Reinforcement Learning (DRL) with Model Predictive Control (MPC).
Unlike traditional MPC methods that rely on static cost function weights, this approach allows real-time adjustment of weights to optimize both trajectory tracking and control smoothness.
The DRL agent learns to adapt to changing environmental conditions, ensuring robust performance even in complex scenarios.
Extensive simulations show that the proposed method achieves superior accuracy in path following, reduces computational load, and enhances overall stability compared to conventional methods.
This adaptive strategy is particularly effective in scenarios requiring high reactivity, making it ideal for real-time autonomous vehicle applications.
Furthermore, the results indicate a significant reduction in trajectory deviation and an improvement in control smoothness, demonstrating the practical viability of integrating DRL with MPC in modern autonomous systems.
Ylona Provot
Title : Localisation de robot - comparaison des méthodes probabilistes et ensemblistes
Reference paper : Sensor Based Robot Localisation and Navigation: Using Interval Analysis and Unscented Kalman Filter.
Immanuel Ashokaraj, Antonios Tsourdos, Peter Silson & Brian A. White
Abstract : La localisation est un des problèmes majeurs en robotique mobile.
Ce travail vise à comparer les méthodes ensemblistes et probabilistes , c'est à dire une méthode de localisation par intervalles ou bien grâce à un filtre de Kalman.
Cet article s'interesse à un problème de localisation dans un environnement constitué de 4 balises auxquelles le robot mesure sa distance.
La résolution d'un tel problème grâce aux intervalles puis à un filtre de Kalman
déterminera laquelle des deux méthodes est la plus efficace dans ce contexte précis en tenant compte des incertitudes des capteurs.
Salah Sekar
Title : Approche Intervalle pour la Localisation de Robots Marins sans GPS ni LiDAR
Reference paper : S. Rohou, B. Desrochers and L. Jaulin (2020). Set-membership state estimation by solving data association, IEEE International Conference on Robotics and Automation (ICRA)
Abstract :
Cet article traite de la localisation d'un robot marin dans un environnement où l'accès aux
informations reçues est limité, dans le cadre d'une situation critique où le GPS, le LiDAR et la
centrale inertielle ne sont pas disponibles. Le robot doit alors se localiser en s'appuyant uniquement
sur une caméra et un Doppler Velocity Log (DVL), en se basant sur des balises préalablement
définies sur une carte. Pour résoudre ce problème, nous proposons une méthode fondée sur le calcul
des intervalles et des contracteurs pour réduire les incertitudes dans les différentes régions
possibles. Cette approche est particulièrement adaptée lorsque la position initiale du robot est
inconnue.
Luc-Andre Terrine
Title : Analyse Comparative des Algorithmes de Planification de Trajectoire pour Robots Mobiles
Reference paper : https://www.mdpi.com/2504-446X/7/3/211
Abstract :
La planification de trajectoire constitue une fonction centrale dans la navigation autonome des robots mobiles.
Cet article examine plusieurs algorithmes de planification de trajectoire, en mettant en avant leurs principes fondamentaux, leurs forces et leurs limites.
Les méthodes abordées incluent les approches basées sur les graphes, les algorithmes heuristiques, l’intelligence artificielle, et les techniques d’évitement d’obstacles.
L’étude évalue la performance de ces algorithmes selon différents critères, tels que la vitesse de calcul, la précision des trajectoires, et leur capacité à s’adapter à des environnements spécifiques.
Main Tihami
Title : DVDS: A deep visual dynamic slam system
Reference paper : Tao Xie, Qihao Sun, Tao Sun, Jinhang Zhang, Kun Dai, Lijun Zhao, Ke Wang, Ruifeng Li,
DVDS: A deep visual dynamic slam system, Expert Systems with Applications, 2025.
Abstract :
Simultaneous Localization and Mapping (SLAM) is critical for autonomous systems and
robotics. This study compares two advanced SLAM frameworks: SLG-SLAM and DVDS.
SLG-SLAM integrates semantic information, laser point clouds, and GNSS data to improve
visual SLAM accuracy, focusing on filtering dynamic targets and refining trajectories
through sensor fusion. Conversely, DVDS leverages learning-based methods, introducing a
dispersive transformer (DisFormer) to exclude dynamic objects and enhance pose estimation
in dynamic, texture-less environments. While SLG-SLAM excels in multimodal data
integration, DVDS demonstrates robustness with deep learning techniques. The comparison
evaluates their accuracy, adaptability, and potential hybridization for advancing SLAM
systems.
Adrian Vanalli
Title : A Study on the Efficiency of a Hybrid Path Planning Algorithm with Dynamic Obstacle Avoidance for Wind-Powered USVs
Reference paper : A Hybrid Path Planning Algorithm for Unmanned Surface Vehicles in Complex Environment With Dynamic Obstacles
Abstract:
Global path planning algorithms have proved to be very useful in targeting autonomous mobility challenges, but equipping fully independent
unmanned vehicles only with these programs lacks robustness and adaptability.
The real world offers dynamic obstacles, and it is therefore necessary to complement the forementioned algorithms with local path planning solutions.
In this study, we will implement the combination of these algorithms for an autonomous sailboat, verify its capabilities on a simulated version of the Brave,
a monohull sailboat developed by students at ENSTA Bretagne, equipped with GPS, a LIDAR, and running on ROS2, and discuss the limitations and implications of this method.
