Preliminary Validation of a COLREGs-Compliant Navigation Framework Using LiDAR and RGB Data Fusion

This paper presents a preliminary validation of a novel navigation framework for Autonomous Surface Vehicles that integrates obstacle detection and path planning while complying with the International Regulations for Preventing Collisions at Sea. While this planning framework was previously tested for rules compliance using assumed obstacle data, this work focuses on testing it with real detection-derived inputs. Obstacle detection combines 3D LiDAR point clouds with YOLOv8-based image interpretation, with an Adaptive Kalman Filter tracking and localizing detected objects for path planning. The planner employs a COLREGs-aware A∗ algorithm, using obstacle positions, velocities, and classes to compute collision-free paths. Validation was conducted in the Stonefish simulator, enabling software-in-the-loop testing with the vehicle's dynamic model.