IIT Madras develops 'SEANAV', a real-time digital twin platform for safer autonomous maritime navigation
Kolkata, Nov 12
Researchers at the Department of Ocean Engineering, Indian Institute of Technology Madras, have developed a real-time digital twin platform designed to test, validate, and train autonomous maritime navigation systems to help vessels navigate safely at sea.
Integrating Artificial Intelligence (AI), Virtual Reality (VR), and multi-sensor fusion, SEANAV constructs a virtual ocean environment where autonomous vessels can be evaluated before setting sail. Built on a Unity-based simulation engine, the platform recreates realistic ocean waves, dynamic shipping traffic, and adverse weather conditions, enabling ships to execute AI-driven collision avoidance and path-following manoeuvres.
The system allows multiple ships to navigate congested waterways simultaneously and offers an immersive VR interface for training maritime professionals. According to the institute, conducting thousands of hours of virtual simulation substantially reduces testing costs while promoting safer and more sustainable shipping operations.
The research has been published in the Journal of Offshore Mechanics and Arctic Engineering (ASME) and presented at ASME OMAE conferences in Vancouver and Tokyo.
Highlighting the importance of environmental perception in dynamic marine environments, researchers noted that existing maritime vision datasets, such as SeaShips and HRSC2016, lack sufficient variation in viewpoints and weather conditions, which limits the real-world reliability of deep learning models.
To overcome this, the team developed a novel custom dataset named 'DetectShips', comprising 14,000 real ship images, and combined it with the SeaShips dataset to train a computer vision algorithm integrated with stereo sensors.
Additionally, the researchers introduced 'StereoLiD', a hybrid sensing framework that combines stereo vision cameras with LiDAR data from the Unity simulation platform. While stereo cameras detect and classify vessels to estimate relative depth, LiDAR inputs further refine depth estimation accuracy. Data from both sensors are fused using a Kalman filter to achieve precise target localisation.
According to the researchers, this multi-vision sensor fusion overcomes common limitations of conventional optical sensors, including fluctuating lighting conditions and dense fog, ensuring dependable target detection and navigation across complex oceanic environments.
— IANS
Leave a comment