DT-Assisted Vehicular Crowdsensing Through Semantic-Aware NDN

Named Data Networking (NDN) has gained momentum in vehicular environments due to its intrinsic features, such as name-based routing and in-network caching, which enhance data retrieval efficiency even under challenging mobility conditions. A key application of vehicular NDN (VNDN) is crowdsensing, where vehicles act as mobile data producers, delivering context-aware information to interested clients. In this context, we propose a Digital Twin (DT)-assisted vehicular crowdsensing framework, in which a DT collects data from vehicles via VNDN. However, the exact name prefix matching required for VNDN packet processing poses challenges in vehicular scenarios, where heterogeneous producers may adopt different naming conventions, making exact matches infeasible. To address this limitation, we propose a semantic-aware packet processing strategy that exploits semantic similarity among content names to enhance data retrieval in VNDN. Our approach integrates neural models for semantic similarity assessment, using the Sentence Transformers architecture, within the VNDN forwarding plane to enable a fallback mechanism when exact matching fails. We assess the feasibility of our solution by testing distinct pre-trained semantic models, used to compute similarity scores between VNDN names, and by varying the similarity thresholds. Experimental results show that incorporating semantic similarity into VNDN forwarding significantly improves data collection performance, enhancing flexibility and interoperability in crowd-sensing scenarios.