Collision Detection and Avoidance for Underwater Vehicles Using Omnidirectional Vision

Area of Science:

• Robotics
• Marine Science
• Computer Vision

Background:

• Underwater exploration relies on image collection, posing risks to vehicles like ROVs and AUVs near the seabed.
• Current obstacle avoidance systems using acoustic sensors are unreliable at close range, demanding high operator attention.
• Advanced mapping, spatial awareness, and safety are crucial for human-operated underwater robotics in confined spaces.

Purpose of the Study:

• To develop an advanced assisted navigation system for ROVs in complex underwater environments.
• To enhance spatial awareness and safety for underwater vehicles operating in confined areas.
• To create a pilot assistance system for inexperienced operators of underwater robotics.

Main Methods:

• Implemented an omnidirectional multi-camera system for a 360° environmental view.
• Developed a visual SLAM framework to create a real-time 3D point cloud of the surroundings.
• Created a risk assessment strategy for nearby obstacles.

Main Results:

• The system successfully generates real-time 360° point clouds of the environment.
• Obstacle risk information is effectively assessed and utilized by the ROV.
• The system enables the robot to perform evasive maneuvers when approaching dangerous obstacles.

Conclusions:

• This novel approach significantly enhances ROV navigation capabilities in complex underwater settings.
• The developed system is a foundational step towards a comprehensive pilot assistance system for underwater robotics.
• It improves safety and operational efficiency for underwater vehicle exploration.