Inria Challenge: improving the detection of targets under water

Imagine being alone in the dark and hearing footsteps in the distance... someone is coming. “In the dark, you can only rely on your hearing to identify where the noise is coming from”, explains Luc Martin, senior R&D consultant at Naval Group. “You don't know how far away the potential threat is. Alternatively, you can use your brain, which has most likely already been confronted with this type of situation, to help you find out more and figure out the situation. Your experience can enable you to assess the direction and distance from which the noise was emitted”.

It's exactly the same situation for a submarine in the depths of the ocean, also submerged in the dark. It can listen in order to assess where sounds are coming from, but it doesn't know precisely how far away they are. To estimate the positions of objects around it, it has only one piece of information: their direction. And one technique: passive trajectography by angle measurement (BOTMA - Bearings-Only Target Motion Analysis). This enables the trajectory of a target to be determined from observations.

“This method assumes that the target maintains a constant speed during observation”, explains Aymeric Bonnaud, Scientific Director at Naval Group. “It involves triangulation, based on angle measurements, which takes into account the movements of objects around the submarine”. However, there are limits to this approach: “The signals are often drowned out by the surrounding noise”, continues Aymeric Bonnaud. “And what can you do when there's only one observer? No triangulation is possible”. As a result, the calculations are imprecise. They sometimes even lead to an infinite number of solutions. So what's the key? Relying on models based on machine learning, i.e. core artificial intelligence techniques, in order to remove uncertainties and establish more reliable positions.

Three partners working together

In this context, the Indian Navy's Defence Research Centre approached Naval Group in 2020. India has long maintained close ties with this industrial group. “The country was the first to be able to build a Naval Group-designed submarine at the first attempt, without having to manufacture the first vessel of its class in France”, says Luc Martin.

In 2021, Naval Group approached two research partners: the Indian Institute of Technology (IIT) in Delhi and Inria, to take up the challenge raised by the Indian Navy in terms of trajectography using passive detection. It has submitted the problem to them in order to explore new ways of improving the technique.

Hybridising control theory and machine learning

The three partners then identified the research teams with the potential to overcome the project’s scientific barriers. On the Indian side, IIT Delhi is one of the country’s leading universities. The chosen researchers, Arpan Chattopadhyay and Arun Kumar, are renowned scientists in the fields of signal processing and machine learning. On the Inria side, the project is led by the Inria Centre at the University of Lille, with the involvement of the Valse (control and cyberphysical systems) and Modal (statistical learning on heterogeneous data) project-teams in Lille, Auctus (robotics) in Bordeaux, Dance (robotics, control and estimation) in Grenoble and Larsen (robotics and artificial intelligence) in Nancy.

“The partnership was officially signed in July 2024, for a period of five years, financed by Naval Group and the Indo-French Centre for the Promotion of Advanced Research (CEFIPRA)”, explains Christophe Biernacki. How do the three partners intend to conduct their research? “Concretely, the approach chosen to improve trajectography by passive detection is to propose different estimation algorithms, by hybridising methods based on classical control theory and artificial intelligence”, explains Christophe Biernacki. Control theory analyses the properties of dynamic systems that can be controlled as a function of an action. This method should make it possible to reduce position uncertainties. 

Artificial intelligence using machine learning will complete the solution: 

In the future, scientists could even have access to real historical submarine data.

Where does the project stand today?

“The last few months have been spent gradually asking the right questions. It's a drawn out process, particularly because it involves a defence issue and because of the cultural differences between France and India”, observes Luc Martin. “But we got there in the end”. There are a number of theses on the horizon: “We started with a use case given by Naval Group and the Indian Naval Defence Research Centre, and the research teams took it on board and reformulated the problem to transform it into subjects of study for five theses and a postdoc, which were submitted to Indian scientists. The theses will be co-supervised by the various entities taking part in this international partnership. One of these research projects is funded by Naval Group. We're in the early stages, but important milestones have been set”.

Progress on this Inria Challenge will be monitored at regular intervals, in addition to the supervision of doctoral students. Improving trajectography is a major challenge for submarine operators, Naval Group's customers. The project could therefore continue beyond the Challenge. To be continued…

• A strategic partnership for Inria and Naval Group, Inria, 10/12/2019.
• Inria Challenges: tackling major issues, Inria, 30/4/2021.
• Poems and Naval Group: working together to model underwater explosions, Inria, 4/6/2021.
• How submarines make themselves invisible and fool sonars (video, in french), National Geographic, 14/12/2022.
• Naval Group reveals the secrets behind the construction of the Barracuda nuclear submarines (video, in french), BFM, 6/10/2023.