TITAN: Temporal-Implicit Topology Attention Network for Resource-Optimal Intelligent Vehicle Collaborative Sensing

Intelligent vehicle collaborative sensing faces critical challenges in dynamic spatio-temporal modeling and resource optimization. Existing approaches suffer from static topological assumptions and lack principled uncertainty quantification for sensing decisions. To address this, we present temporal-implicit topology attention network (TITAN)-strategic multicriteria active resource targeting (SMART), a unified framework integrating the TITAN with SMART. TITAN incorporates three innovations: implicit topology relation learner (ITRL), dynamic attention neighborhood aggregator (DANA), and temporal propagation attention module (TPAM). Importantly, TITAN is a spatio-temporal attention network, where ITRL and DANA operate on implicit and explicit spatial relationships, while TPAM captures multiscale temporal propagation dynamics. Experimental validation demonstrates that TITAN–SMART achieves target accuracy with up to 46.2% less resource cost than state-of-the-art baselines, maintaining scalability across large-scale networks. By explicitly targeting the gap between passive prediction on fully observed graphs and active sensing under strict sensing-resource constraints, TITAN–SMART enables principled selection of where to sense next rather than only forecasting future states. The framework’s interpretable decision patterns establish it as a practical solution for industrial Internet of Things (IoT) deployments.