Communication-Aware Autonomous Underwater Vehicle Framework with Multimodal Temporal Transformer for Integrated Perception,Routing,and Predictive Maintenance

The growing applicability of Autonomous Underwater Vehicles (AUVs) in oceanographic studies, environmental monitoring, and infrastructure inspections, and infrastructure inspections is limited by the problem of underwater acoustic communication, which includes low bandwidth, high propagation delay, and high-power limitations. Current AUV networks are based on some form of static routing and reactive fault management, which are not suitable to large-scale, long-range underwater missions. This paper introduces a new communication-aware architecture for AUV networks which is based on a multimodal temporal transformer (MMTT)-based framework. In this framework, communication, routing and predictive maintenance are considered a multi-task sequential learning problem, to be optimized as a joint problem on a common temporal representation. It is based on a time-varying dynamic graph model, and it considers the mobility of nodes, environmental perturbations, and varying channel conditions. Evaluations of representative underwater scenarios through simulations show that our solution is improves packet delivery, energy savings, and route efficiency by factors of 20, 15, and 30, respectively than baseline protocols. Our framework is also able to reduce end-to-end latency by a quarter, ensuring timely data delivery in a dynamic underwater setting. These advances allow the proposed system to be more reliable and energy-efficient thereby supporting long-duration, scalable, and autonomous underwater missions.