PhD thesis defense on “Smart River: Towards Efficient Cooperative Autonomous Inland Navigation“

PhD Committee:

• Toufik AHMED Professeur, Bordeaux INP, Rapporteur
• Thierry DELOT Professeur, Université Polytechnique Hauts-de-France, Rapporteur
• Franck GECHTER Maître de conférences (HDR), UTBM, Examiner
• Pietro MANZONI Professeur à l’Université Polytechnique de Valence, Espagne, Examiner
• Charlotte LOGEAIS Chargée de mission bâteau autonome, VNF, Examiner
• Yvon GOURHANT Ingénieur de recherche, Orange Labs, Examiner
• Sidi-Mohammed SENOUCI Professeur, Université de Bourgogne, Supervisor
• Philippe BRUNET Maître de conférences, Université de Bourgogne Co- Supervisor
• Metzli RAMIREZ-MARTINEZ Enseignante chercheure, Université de Bourgogne, Invited

Abstract:

In recent years, inland waterway transport has witnessed increasing attention from France and many European countries. However, this mode of transport lacks flexibility, has an aging infrastructure and the current ships are not adapted to an increase in transport capacity ensuring the safety of vessels and goods as well as reliable and constant delivery times. Therefore, inland transport must go through an organizational and technical renovation specific to its particular environment in order to hope to compete with land transport. In this thesis, we propose developing a smart river ecosystem that focuses on three principal axes: (i) automatic inland infrastructure, (ii) autonomous inland ship, and (iii) promoting connected and cooperative navigation. The first axis focuses on efficiently automating the existing inland infrastructure using the Lock Automation Decision Making (Lock-ADM). The Lock-ADM algorithm works in three steps. First, it calculates the optimal number of locks to automate for a given investment cost. Then it measures the importance of the locks in the network according to several criteria. Finally, it selects the best locks to automate using a genetic algorithm. Lock-ADM allows the annual planning of the locks to be automated, starting with the most constraining ones for the current network. The second axis focuses on the development of an environment perception system for autonomous ships. It allows to delimit the navigable zones where a ship can navigate safely and thus avoid any obstacle on its way. To do so, we have built the first open-source dataset (labeled images) for the river domain: InlandAutoDetect. We have exhaustively labeled the different objects that make up the river navigation. Then, we compared the performances of nine deep learning algorithms in terms of detection accuracy and response time. We selected the Retinanet algorithm which showed the best performance to delimit with accuracy and in real time a safe navigation zone for our autonomous ship. Finally, the third axis introduces C-IAShips, an architecture based on Blockchain and MEC (Mobile Edge Computing) technologies for cooperative autonomous ships. The proposed architecture guarantees low latency and efficient communication while protecting the confidentiality of the ships and the security of the exchanged data. The main advantage of cooperative ships is that they allow for a more powerful and efficient operation of the overall system. In particular, we have studied the feasibility of two cooperative applications: the first for scheduling the passage of ships at locks and the second for collision detection.

Keywords: 

Autonomous Ships, Internet of Ships, Cooperative Intelligent Transportation Systems, Automated locks, Connectivity, Machine Learning