This article is devoted to the development of a situational awareness system architecture for a Level 4 autonomous marine vessel.

   The main problem addressed in this work is the inability of classical data fusion algorithms (Kalman filters, neural network-based fusion) to distinguish between sensor failures, accuracy degradation, and targeted cyberattacks, which is critical for ensuring safe unmanned navigation of an autonomous vessel.

   The proposed architecture includes three key components. The first component is a two-tier sensor organization: a continuously active core consisting of primary navigation and specialized sensors (LiDAR, optical cameras, and thermal imagers), activated according to the scenario. To comply with Rule 5 of the International Regulations for Preventing Collisions at Sea (COLREGs-72) concerning continuous lookout, the concept of a “non-switchable sensor core” is proposed — a minimum set of sensors that must function continuously regardless of the scenario. Adaptation to navigation conditions is achieved not by disabling sensors, but by dynamically weighting their data in the data fusion module. The second component, the Sensor Trust Manager, implements a sensor connection management protocol that ensures the isolation of unreliable sources before the data enters the fusion process. The third component, the navigation scenario classifier, dynamically assigns confidence weights to sensors in covariance-based fusion algorithms based on illumination, visibility, traffic density, and navigation area.

   The novelty of the proposed approach lies in the introduction of meta-awareness: the situational awareness system evaluates not only the external environment but also its own state (“sensor trust”).

   For the first time, a transition from the “data fusion” paradigm to the “source trust management” paradigm is proposed, creating a technological foundation for the design of autonomous vessels.