SDVDiag: Multimodal Causal Discovery for Online Diagnosis in Software-defined Vehicles
A new research paper introduces SDVDiag, a multimodal causal discovery pipeline designed for diagnosing issues in software-defined vehicles. This system fuses log-based and metric-based data into a shared embedding space to construct causal graphs, and it operates in an online, anomaly-driven mode rather than offline. Evaluations on an Autonomous Valet Parking testbed demonstrated that SDVDiag produces sparser causal graphs and outperforms a metrics-only baseline in accuracy, even recovering root causes several hops away from observable symptoms. AI