Faites connaître cette offre !
Reference : UMR5506-MARGOU-004
Workplace : MONTPELLIER
Date of publication : Wednesday, June 9, 2021
Scientific Responsible name : Marc GOUTTEFARDE
Type of Contract : PhD Student contract / Thesis offer
Contract Period : 36 months
Start date of the thesis : 1 October 2021
Proportion of work : Full time
Remuneration : 2 135,00 € gross monthly
Description of the thesis topic
Ph.D. Position, LIRMM (CNRS/University of Montpellier, France) - Machine Learning approaches for litter detection and classification in underwater videos for automated robotic seabed cleaning.
--- Context ---
The global marine plastic litter challenge comprises an estimated stock of 83 million tons of plastic waste accumulated in the oceans. The recovery of plastic materials already disseminated worldwide in the sea is an arduous and costly task. Therefore, innovations are urgently needed.
The EU-funded MAELSTROM project (https://www.maelstrom-h2020.eu/) is bringing together key stakeholders – from research centers and recycling companies to marine scientists and robotic experts – to leverage the integration of complementary technologies for the sustainable removal of marine litter in different European coastal ecosystems. The MAELSTROM project will design, manufacture, and integrate scalable, replicable, and automated technologies to identify, remove, sort and recycle all types of collected marine litter into valuable raw materials.
Existing solutions for marine litter removal on the seabed are considered either highly harmful for the marine ecosystem (dredges or bottom trawl nets) or inefficient (ROV, divers). MAELSTROM seabed cleaning solution is a cable-based underwater robot installed and operated from a floating barge and equipped with selective cleaning tools (see Figure 1). The operator selects and install the desired end-effector (cleaning tool) on the cable robot and starts performing the cleaning of the area, either by manual teleoperation (remote operation), by shared control (i.e., computer-aided teleoperation) or semi-autonomously (i.e., by programming the robot to execute autonomously a specific task or trajectory). The goal is the selective removal of the litter, thus minimizing the environmental impact. Once the area is cleared, the platform will be moved to the next area of interest.
The LIRMM laboratory ( www.lirmm.fr ) is involved in the design of the cable-driven underwater robot and in the development of Machine Learning algorithms to detect the litter. This last application will be the research topic of the Ph.D. thesis.
--- Objectives ---
The objective of the Ph.D. thesis is to build a system able to provide suggestions - in the form of areas/objects highlighted in the interactive image seen by the robot operator - regarding which regions or objects on the seafloor should be marked for removal. The system will use Artificial Intelligence machine learning techniques such as Deep Learning or Reinforcement Learning. These algorithms will allow the robot to learn from the previous decisions made by the operator during the shared control mode.
A dataset for training of the machine learning algorithm will be generated based on the human operator work which clicks on the images to select the litter to be collected by the robot. The machine learning algorithm will then continuously improve its ability to identify potential targets and areas of interest. In all cases, the decision to clean a target will be on the operator's responsibility to deal with the preservation of potential archaeological artefacts or to avoid any contact with submersed ammunitions or dangerous chemicals.
The thesis will be supervised mainly by researchers specialized in Deep Learning methods (Marc Chaumont http://www.lirmm.fr/~chaumont/ and Gérard Subsol https://www.lirmm.fr/~subsol/ ), assisted by researchers specialized in cable and underwater robots (Marc Gouttefarde https://www.lirmm.fr/users/utilisateurs-lirmm/marc-gouttefarde and Vincent Creuze https://www.lirmm.fr/users/utilisateurs-lirmm/vincent-creuze ).
--- Required competences ---
• Master of Science or equivalent in robotics or computer vision
• Strong knowledge in programming (C++, Python)
• Strong background in Deep Neural Network frameworks (e.g., PyTorch, TensorFlow…)
• Already worked on project requiring image detection by Deep Learning
• Background in image/signal processing libraries such as OpenCV
• Awareness of 3D reconstruction and 3D visualization would be a plus.
• Get used to handling reproducibility tools (Markdown, GitHub, Docker)
• Rigorous, autonomous, creative
• Fluent English speaker to be able to interact with the European partners of the project.
--- Duration, location, and salary ---
• Duration: 36 months
• Location: LIRMM laboratory, Montpellier (France)
• Gross salary per month: 1,982€ (year 1 & 2) and 2,085 € (year 3)
Applications should be submitted at your earliest convenience and should contain at least:
• CV, including previous experience relevant to the project
• A PDF file containing a motivation letter and the transcripts of grades of the Master of Science
This project has received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No 101000832 --- Recrutement financé par le programme de recherche et d'innovation H2020 MAELSTROM de l'Union Européenne, GA n°101000832.
The Laboratoire d'Informatique, de Robotique et de Microélectronique de Montpellier (Laboratory of Computer Science, Robotics and Microelectronics of Montpellier) is a major multi-disciplinary French research center located in the South of France. It is affiliated with the University of Montpellier and the French National Center for Scientific Research (Centre National de la Recherche Scientifique, CNRS).
It conducts research in Computer Science, Microelectronics and Robotics and is organized along 3 departments comprising 19 international research teams assisted by central services personnel.
Constraints and risks
No particular constraint or risk
We talk about it on Twitter!