Data-Driven Situational Awareness and Collision Avoidance Framework for Autonomous Surface Vehicles

Abstract

The rapid development of autonomous surface vehicles (ASVs) offers significant potential to improve maritime safety, operational efficiency, and sustainability. This review examines data-driven frameworks for situational awareness (SA) and collision avoidance (CA) in ASVs, conceptualizing them as representative intelligent engineering systems (IES) empowered by artificial intelligence (AI). Anchored in cyber-physical systems (CPS) theory and information theory, the framework combines multi-modal sensor fusion, machine learning for perception and trajectory prediction, and reinforcement learning (RL) for real-time decision-making. The paper analyzes four core transformation pathways enabled by AI: from standardized rule-based navigation to personalized adaptive strategies, from reactive maneuvers to proactive prediction and optimization, from manual or remote operation to autonomous intelligent control, and from isolated platforms to integrated cyber-physical maritime ecosystems. Specific examples include deep reinforcement learning (DRL) agents that achieve COLREGs-compliant avoidance and digital-twin-supported predictive maintenance for propulsion systems. While these approaches deliver measurable improvements in safety metrics and efficiency, the review critically evaluates persistent challenges related to data quality under adverse marine conditions, regulatory integration, computational constraints, and cybersecurity risks. Future directions emphasize hybrid neuro-symbolic methods, physics-informed models, and multi-agent coordination to support trustworthy deployment. The work provides a structured foundation for advancing certifiable autonomy within broader IES paradigms.