Final tests in Lake Starnberg: EU partners use generative AI to improve the environmental perception of underwater robots

Whether in murky waters, cramped ship tanks or caves with little light – poor visibility conditions have a huge impact on the environmental perception of autonomous robots. How can they still operate safely and reliably and perform complex tasks?

The diving platform on Lake Sternberg (Dirk Osmers, DFKI)

DFKI-AUV Flatfish in the water (DFKI, Dirk Osmers)

A diver from the THW enters the water (DFKI, Tom Becker)

DFKI researchers control the AUV and analyze the sensor data (DFKI, Tom Becker)

A monitor shows the data input and the AI-generated output (DFKI, Tom Becker)

In the project DeeperSense, coordinated by the German Research Center for Artificial Intelligence (DFKI) since January 2021, an international consortium pursued a completely new approach in underwater robotics: Through Artificial Intelligence, especially Deep Learning, it combined the strengths of visual and acoustic sensors to significantly improve the perception of robotic systems under challenging conditions. Specifically, the goal was to generate high-resolution camera-like data from low-resolution sonar data in three different use cases from the maritime domain.

In Use-Case 1 the Robotics Innovation Center collaborated with Kraken Robotics GmbH and the Technical Assistance Service (THW). An autonomous underwater vehicle (AUV) from DFKI was deployed to monitor THW divers at work underwater. Typically, a remotely controlled robot with a camera handles a task like this, ensuring the safety of divers by transmitting its data to a control station. However, optical sensors underwater are of limited use, so the project partners equipped the AUV with an additional sonar sensor. Using AI algorithms, the researchers succeeded in teaching the acoustic sensor to provide camera-like images that can be easily interpreted by human personnel.

At the end of the project, the partners gathered from December 10 to 16, 2023, at Lake Starnberg. In the fifth-largest lake in Germany, they successfully demonstrated the functionality of their technologies under real conditions based on Use-Case 1.

Other partners in the DeeperSense project included the University of Girona, the University of Haifa, the Israel Nature and Parks Authority, and Tecnoambiente SL. The project was funded under the EU Horizon 2020 research framework program.

For more information:
Project DeeperSense

Dr. Thomas Vögele

Further News

last updated 28.05.2024
to top