The Smart Ports Chair awards a prize to an autonomous robotics system with monocular vision for the maintenance of underwater infrastructure

 

La Cátedra Smart Ports, driven by the Universitat Jaume I and Autoridad Portuaria de Castellón, has awarded Inés Pérez Edo with the 3rd Prize for Best Final Degree Project. The project develops a technology of autonomous underwater robotics capable of inspecting and repairing port pipelines using machine vision, reducing costs and occupational risks.

 

A qualitative leap in the maintenance of submerged infrastructure

 

The management of underwater assets, such as outfalls, pipelines, and foundations, has historically presented a logistical and safety challenge for port operators. Traditionally, these tasks rely on specialized divers, which entails elevated human risks, significant operating costs and technical limitations resulting from low underwater visibility.

 

In response to this problem, the research of the student of the Degree in Robotics Intelligence of the Universitat Jaume I (UJI) introduces a paradigm shift: the use of autonomous robots that operate without direct human intervention. This advance aligns with the strategy of Ports 4.0looking for more infrastructure connected and resilient.

 

Monocular vision technology and Deep Learning

 

The core of the innovation lies in its perception system. The marine environment presents severe obstacles for traditional sensors due to light absorption and turbidity. Unlike expensive solutions that require multiple laser sensors, the award-winning proposal uses monocular visionThis allows the robot to interpret its environment using a single standard camera.

 

To achieve this autonomy, algorithms have been implemented Deep Learning, specifically the model YOLOv8 (You Only Look OnceThis technology allows for the detection and segmentation of pipes in real time with remarkable accuracy, even in reduced visibility conditions, calculating the approach trajectory and orientation for a secure grip.

 

Key project data and technical results

 

The specifications and results obtained during the validation phase of the project are detailed below:

 

Parameter	Detail / Result
Organizing Entity	Cátedra Smart Ports (backed by Puertos del Estado)
Basic Technology	Robotized BlueROV2 modified + AI Model YOLOv8
Detection Reliability	Superior to 90 % in simulated and real-world environments
Real Test Dataset	55 previously unseen images of Puerto de Castellón
Prize Endowment	1.000 Euros

 

Real-world validation: From the laboratory to the Port of Castellón

 

One of the most relevant milestones valued by the awards committee has been the system's ability to transcend the laboratory environment of CIRTESU (Center for Research in Robotics and Underwater Technologies). The field tests were carried out in the waters of the Puerto de Castellón.

 

During these tests, the algorithm demonstrated a high capacity for generalization by identifying pipes using a dataset it had not encountered during its training. This confirms that the technology is robust against variability of a royal harbor basin, where currents and lighting are constantly changing.

 

• Autonomous detection: Asset identification without human intervention.
• Dynamic navigation: Tracking complex trajectories for approach.
• Precise manipulation: Execution of gripping maneuvers, a preliminary step for repairs or assemblies.

 

The relevance of the work of Inés Pérez Edo This has also become evident in the academic field, with some of its findings being presented at the specialized forum Automar from the Automation Conference.

 

Key points and frequently asked questions about Autonomous Port Robotics

 

What advantages does autonomous robotics offer compared to traditional divers?
Primarily, it eliminates life-threatening risks to human operators in hazardous environments and significantly reduces operating costs, enabling more frequent and accurate inspections without relying on human visibility.

 

What is monocular vision applied to underwater robotics?
It is an efficient technology that allows the robot to "understand" and navigate its environment using a single standard camera, processing the images through Artificial Intelligence (YOLOv8) instead of relying on expensive multi-sensor or laser systems.

 

Has this technology been tested in real-world environments?
Yes. The project has been successfully validated in the waters of Puerto de Castellóndemonstrating that the algorithm works correctly with previously unseen real images, overcoming the limitations of controlled laboratory environments.