Annonce

Les commentaires sont clos.

Postdoc position - Ship Motion Prediction by Deep Learning

26 Janvier 2022


Catégorie : Post-doctorant


Postdoc position - Ship Motion Prediction by Deep Learning

 
Starting May 2022; 12 months, renewable. This position is limited to applicants of European citizenship.
 
Keywords
Artificial intelligence; deep learning; Time series; Motion prediction; Ship Stabilization
 
Context
The overall objective of this work is to participate in securing naval operations of various kinds with regard to the state of the sea, in particular by anticipating potentially dangerous ship movements to be able to take corrective measures ( stabilization) or preventive measures (interruption of tasks that could be impacted).
 
Scientific objectives
Many existing and operational systems already make it possible to stabilize ships. These systems are generally based on movement measurements using inertial units, combined with signal processing techniques that make it possible to predict movements a few seconds in advance. We aim to improve the movement prediction horizon, aiming for a few tens of seconds, in order to be able to guarantee the safety of operations in heavier seas. Increasing this prediction time could initially involve the use of machine learning techniques for inertial data processing [ZHAO, 2004]. Still, it can only really progress by considering additional context information, mainly through vision, which makes it possible to observe the state of the sea around the ship to anticipate the arrival of significant movements.
 
There are already many works on the measurement of the state of the sea (e.g., [BENETAZZO, 2018]) focused on the measurement of the surface, which can be exploited in this problem by being associated with motion prediction models of the surface and a ship motion model on this surface. However, we wish to study whether an end-to-end model, starting from the vision of the sea (associated with inertial measurements) to give the ship motion prediction directly could be more effective.
 
More specifically, the objectives of this work are:
• Increase the prediction horizon of a ship's movements by exploiting artificial intelligence techniques
• Go beyond possible predictions using an inertial unit and conventional signal processing techniques through the use of deep learning
• Exploit visual information to anticipate longer-term movements by observing incident waves
• Use 'end-to-end' techniques to reduce the composition of errors compared to methods that break down the problem (wave detection, wave motion prediction, ship motion prediction).
 
 
Description of main activities
The role of the post-doctoral student in this project will be to participate in the following four tasks in collaboration with engineers from ENSTA Paris and Naval Group - Sirehna:
 
• Task 1: Detailed definition of the problem (fine selection of quantities to be predicted (roll, pitch, complete movement, etc.), choice of AI techniques to be implemented (recurrent networks, convolutional networks, transformers, etc.), detailed specifications of the work of tasks 2,3 and 4.
• Task 2: Creation of a database of IMU/vision sequences from a photo-realistic simulator for different ships, different sea states, different sensor configurations.
• Task 3: Development of a motion prediction method from inertial measurements, comparison with conventional approaches (on simulated and real data).
• Task 4: Integration of video data in the motion prediction (on simulated data), evaluation of the gain compared to the previous approach.
 
Knowledge or skills for the position
• Experience in Deep Learning with the standard libraries of the domain (pytorch, tensorflow, …)
• Experience in control or in time series prediction
• Good development experience (mainly python) and associated tools (git, test methods, integration, …)
• Curiosity, autonomy, capacity for synthesis
 
Profile and/or desired experience
Required: PhD in the field of deep learning, vision, or estimation
Desired: Development experience in an industrial context
 
Locations
The position will be mainly located at ENSTA Paris (828 boulevard des maréchaux, 91120 PALAISEAU), with collaboration and missions at Naval Group - Sirehna (Technocampus Ocean - 5 rue de l'Halbrane - 44340 Bouguenais)
 
Contacts and application
David Filliat - ENSTA Paris
david.filliat@ensta-paris.fr
 
 
--- This position is limited to applicants of European citizenship ---
 
Please include in your application within one pdf-file sent by e-mail to david.filliat@ensta-paris.fr :
− Cover letter outlining (i) how you meet the requirements for the position, (ii) relevant details of your past research projects explaining how your previous experience is adapted to this project.
− Curriculum vitae (including publication record: journal papers, conference contributions, etc.)
 
Bibliographical references
[BENETAZZO, 2018] Benetazzo, A., Serafino, F., Bergamasco, F., Ludeno, G., Ardhuin, F., Sutherland, P., ... & Barbariol, F. (2018). Stereo imaging and X-band radar wave data fusion: An assessment. Ocean Engineering, 152, 346-352.
[ZHAO, 2004] George Zhao, Roger Xu, and Chiman Kwan. “Ship-motion prediction: Algorithms and simulation results”. In: vol. 5. June 2004, pp. V–125. isbn: 0-7803-8484-9