Charles COQUET, PhD student with Thales DMS, supervised by Benoit ZERR (ENSTA Bretagne), Pierre-Jean BOUVET (ISEN) and Andreas ARNOLD-BOS (Thales DMS)
"Control of a fleeing robotic swarm formation to track a dynamic target with communication constraints: analysis and simulation"
We describe and analyse Local Charged Particle Swarm Optimization (LCPSO), an algorithm we designed to solve the problem of tracking a moving target releasing scalar information in the environment, using a swarm of agents. This method is inspired by flocking algorithms and the Particle Swarm Optimization (PSO) algorithm for function optimisation. Four parameters drive LCPSO: the number of agents ; the inertia weight ; the attraction/repulsion weight ; and the inter-agent distance . Despite its simplicity, LCPSO has many interesting properties, which we demonstrate and show in this paper. Using Artificial Potential Fields (APFs), we provide a mathematical analysis of the LCPSO algorithm under some simplifying assumptions. By focusing on a sample application of target tracking with communication constraints, we then remove those assumptions one by one. Firstly, we show the algorithm is resilient to constraints on the communication range, and the behaviour of the target. Secondly, we show there can be no flocking behaviour when the number of agents becomes too large. Thirdly, we demonstrate that the swarm’s speed is limited intrinsically by its parameters. Finally, we demonstrate that a fleeing agent emerges in the swarm, which is intrinsically seen as a leader in this multi-agent system. Results on simulation then confirm the theoretical analysis. This provides useful guidelines to understand and control the LCPSO algorithm as a function of swarm characteristics as well as the nature of the target.
Yoann SOLA, PhD student with ENSTA Bretagne, supervised by Benoit CLEMENT and Gilles LE CHENADEC
"Deep Reinforcement Learning for the Control of AUVs with Stability Guarantees"
In this presentation, we will see how we can control an Autonomous Underwater Vehicle with Deep Reinforcement Learning algorithms in order to make it adaptive to its environment. Moreover we will see how we can give stability guarantees of the control with the use of Lyapunov Neural Networks. Finally, we will see the tools allowing to simulate the AUV with the presentation of the UUV Simulator.
Thibaut NICO, PhD student at ENSTA Bretagne and ECA Robotics, supervised by B. Zerr, L. Jaulin (ENSTA Bretagne) and S. Tauvry, H. Ott (ECA Robotics)
"Study and development of relocation solutions of underwater objects by heterogeneous underwater vehicles"
In the Mine Counter Measure (MCM) context in the underwater environment, it is vital to revisit some potentially dangerous objects to identify and neutralize them if they are actually mines. This dangerous task was usually performed by human divers but more and more it is conducted by unmanned underwater robots. Due to the low cost design of the revisit/mine-killer robot, going straightforward to the geolocalized suspicious object does not guarantee that the robot will redetect it. Moreover the robot may dive at a far position from the target and the lack of absolute positioning system in underwater environment demands a strategy to follow to guarantee the revisit of this target. Based on a priori information in the working area and especially the presence of geolocalized landmarks, the problem is solved as a motion planning problem considering uncertainties due to the increasing error when navigating underwater.
In the context of bounded errors, the problem is solved in a set-membership manner.
Firstly, based on the location and the shape of the landmarks, and on the visibility area of the sensor embedded, the registration maps are computed indicating the sets of robot poses to detect the different landmarks considered in order to reduce the uncertainty on the robot position. Secondly, based on a parametric motion model with uncertain parameters, an high level strategy is provided through a graph optimization. The strategy consists in navigating between the registration maps to reduce each time the uncertainty in position of the robot and finally to guarantee the reachability of a goal area corresponding to the redetection of the target.
Fabien NOVELLA, Marie PONCHART et Pierre BENET, Research engineer at ENSTA Bretagne - NAVIDRO Project
"State-of-the-Art of Standalone Accurate AUV Positioning - Application to High Resolution Bathymetric Surveys"
Irène MOPIN, Research engineer at ENSTA Bretagne - S2MF Astrid Project
"Use of non-linear acoustic properties of sea-water to characterize the seafloor. Multifrequency singlebeam echosounder and side scan sonar design and development"
Singlebeam echosounders and sonars are nowadays mainly used to study the watercolumn (fisheries acoustics) or the seabed by associating the acoustic backscattering energy to the sounding. In the watercolumn, scatterers (fishes, plancton layers, …) acoustic responses are typical of each targets and their frequency variations allow us to identify species . Likewise, each type of seafloor has its particular backscattering index (BS for Backscattering Strength) which depends on frequency f and also on beam incident angle θ on the seabed . The resulting BS(f,θ) curves are used for seafloor characterisation. In that respect, to study all those targets by acoustics, a large panel of frequencies is necessary. But, as sounders of the market are mostly narrow-band, several echosounders must be used in parallel to achieve this aim. Such a system becomes quickly bulky and costly when a large range of frequencies is needed. Therefore, we propose in the S2MF project two prototypes of singlebeam echosounder and sonar able to generate multiple frequencies simultaneously thanks to non-linear properties of acoustic waves in sea water. Wide-band receivers had also been developed in this project with the purpose to characterise the seabed. First, the principles of waves generation and their non-linear propagation will be exposed, with associated in-tank measurements and model. Then, the reflectivity data-processing will be explained : computation of the backscattering index BS, sonar equation compensations, estimation and modelling of BS(f,θ) curves. Finally, results and analysis of measurements at sea will be discussed.
 Laurent Berger, Verena M. Trenkela. A fisheries acoustic multi-frequency indicator to inform on large scale spatial patterns of aquatic pelagic ecosystems. Ecological Indicators, Integrating Sciences for Monitoring, Assessment and Management, 2013
 Xavier Lurton, Ifremer, GEOHAB Workshop, 2013
Thanh Huy NGUYEN, PhD student at IMT-A Plouzané, supervised by J.M. Le Caillec and D. Gueriot
"Methodology of 3D mosaic construction by the fusion of airborne LiDAR data and optical imagery"
The perception of an environment and its follow-up applications require using multiple heterogeneous sensors to capture specific and informatively complementary characteristics of this environment, in order to improve the scope and/or the quality of the acquired information . In an urban and natural context, airborne LiDAR (Light Detection And Ranging) systems are widely used for providing accurate 3D surface information and 3D geometry of objects and ground elements in the scattered-point data modality thanks to its range detection principle; whereas aerial and satellite photogrammetry supplies semantic and texture information in the form of spectral imagery. Over the years, existing works in the domain of data fusion between optical imagery and airborne LiDAR data have addressed very specific acquisition contexts, in which the datasets are already registered and/or they are acquired from the same platform, at identical or very close dates  . As a consequent, they have never intended to overcome the inherent obstacles of the context where data sets collected from different platforms with different acquisition configuration (e.g. flying track, height, orientation, etc.) at different moments and even in different seasons, with different spatial resolutions and levels of detail. This context also relates to the rise of Geographical Information System (GIS) availability, in particular through the open data movement, that requires the integration of data from multiple and heterogeneous sources. However, a solution that is versatile enough to satisfy this difficult context has not been found yet . Motivated by this unresolved research problem, we propose a methodology of fusion of airborne LiDAR data and optical imagery, including a novel feature-based registration approach capable of overcoming the challenges of the aforementioned research context. We focus on urban scenes and more specifically on buildings as primitives on which the matching between the datasets relies.
 Mitchell, H. B. (2007). Multi-Sensor Data Fusion: An Introduction. Berlin, Heidelberg: Springer Berlin Heidelberg.
 Debes, C., Merentitis, A., Heremans, R., Hahn, J., Frangiadakis, N., van Kasteren, T., Pacifici, F. (2014). Hyperspectral and LiDAR Data Fusion: Outcome of the 2013 GRSS Data Fusion Contest. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 7(6), 2405–2418.
 Vo, A.-V., Truong-Hong, L., Laefer, D. F., Tiede, D., D’Oleire-Oltmanns, S., Baraldi, A., … Tuia, D. (2016). Processing of Extremely High Resolution LiDAR and RGB Data: Outcome of the 2015 IEEE GRSS Data Fusion Contest—Part B: 3-D Contest. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 9(12), 5560–5575.
 Zhang, J., & Lin, X. (2017). Advances in fusion of optical imagery and LiDAR point cloud applied to photogrammetry and remote sensing. International Journal of Image and Data Fusion, 8(1), 1–31.
Title : Blackbox and Gray-box Optimization
Speaker : Jordan Ninin, associate professor, Lab-STICC, ENSTA Bretagne
Abstract: Some problems do not possess the required structure to be addressed by classical optimization methods: find the best default parameter of software, or minimize a function build from a numerical simulation. In this presentation we focus on optimization problem where the function to minimize come from an expensive simulation in terms of computational time. This problem can be non-smooth, discontinuous and may even be contaminated by numerical noise. There are named blackbox optimization or derivative-free optimization problems. We also detail an idea to deal with problems where the objective function or constraints are partially known (gray-box optimization), by combining two softwares NOMAD and IBEX.
Title : Satellite SAR interferometry
Speaker : Jean-Marie Nicolas, Professor at IDS (Images, Données, Signal) department, Telecom ParisTechDownload : confinterferometrie_jmnicolas_270618.pdf (6.94 Mo)
Title : Locomotion dynamics for bioinspired robotics
Speaker : Frédéric Boyer, Professor in robotics, LS2N, IMT-Atlantique, Nantes
The second part of the day is dedicated to scientific talks related to the thesis. The following presentations are accessible to a wide audience.
The localization of underwater robots remains a challenging issue. Usual sensors, such as Global Navigation Satellite System (GNSS) receivers, cannot be used under the surface and other inertial systems suffer from a strong integration drift. On top of that, the seabed is generally uniform and unstructured, making it difficult to apply usual Simultaneous Localization and Mapping (SLAM) methods to perform a localization.
Hence, innovative approaches have to be studied. The presented method can be characterized as a raw-data SLAM approach, but we propose a temporal resolution — which differs from usual methods — by considering time as a standard variable to be estimated. This concept raises new opportunities for state estimation, under-exploited until now. However, such temporal resolution is not straightforward and requires a set of theoretical tools in order to achieve the main purpose of localization.
This thesis is thus not only a contribution in the field of mobile robotics, it also offers new perspectives in the areas of constraint programming and set-membership approaches. We provide a reliable contractor programming framework in order to build solvers for dynamical systems. This set of tools is illustrated along this work with realistic robotics applications.
Keywords: mobile robotics, dynamical systems, constraint programming, interval analysis, localisation, SLAM, AUVs
Guillaume Sicot, Associate Professor, ENSTA Bretagne
"On the accuracy of the estimated parameters from airborne hyperspectral data on the coastal zone"
Page(s): 1 2