UFR Sc Techniques

The objective of this lecture is to provide insight specific to deep-sea underwater vehicles and their operation. It gives an overview of existing underwater vehicles, observatories and operations procedures, with emphasis on the Underwater Systems developed and operated within the French Oceanographic Fleet. Sensors and equipment required for deep water submarine operations will be presented. A focus on acoustic communication necessary for the operation of AUVs (Autonomous Underwater Vehicles) will be carried out in terms of principle, strategies and equipment. Instrumental architecture and key principles of distributed multi-sensor acquisition systems will be exposed. Visual perception in deep-sea operations as well as 3D representation of natural structures on the ocean floor will also be deepened

This module aims at equipping students with the ability of creating machine learning (and deep learning) models for usual intelligent robotic perception tasks. Computer vision and machine listening on a marine environment are the two target fields of application of the contents of this course, however other modalities (e.g. hyperspectral imaging, natural language,...) will be discussed. After the module students will be capable at identifying perception tasks to which deep learning techniques can be applied. Students will be able to build specific models to solve these problems and assess their performance. They will also know the main specificities of image and audio for marine applications.

Participants will carry out tutorial and practical work on marine and underwater robotsunder the supervision of lecturers. Matlab/Simulink and Ubuntu/ROS will be used to take over the following robotic platforms :• Marius sailing robot• 8-thruster BlueROV underwater robot• 3-thruster fully equipped CORAL underwater robot

This course presents the main framework of reinforcement learning (RL). At the end of the course, the student will be able to identify and frame problems under this formalism. He will master the vocabulary associated with the RL field, and will be aware of the main obstacles to tackle when using and developing RL approaches (e.g. exploitation-exploration trade-off, sample efficiency). Moreover students will be capable of understanding and implementing “vanilla” algorithms of the core methods in RL (e.g. Q-learning and policy optimisation) and analyze/apply some specific advanced algorithms (e.g. A2C and PPO).

Module servant à recueillir tous les commentaires des étudiants du MIR, depuis les évaluations des matières jusqu'aux choix des options d'études.

After this course the students will be able to identify the deep learning (DL) approaches to be applied to multiple types of machine learning problems, depending on the task and the data inputs/outputs. Students will know how to build and train advanced DL models with the use of existing publicly available software tools. We will also introduce the main shortcomings and limitations of deep learning techniques such as the problem of interpretation and the exploitation of adversarial attacks.

Objectives:The aim of this course is to give the students understanding of both Physics of acoustic waves and signal processing techniques, in order to be able to suggest which system is appropriate for given specifications, to predict its performances through simple models, to implement data processing algorithms and to interpret the results. The complementarity of deep learning techniques with physics based approach is highlighted.Contents:- Propagation of acoustic waves in shallow and deep water, scattering, Doppler effect.- Signal processing : positioning, detection and ranging, noise correlation, time reversal.