PAMPOS: Causal Transformer Detects V2X Misbehavior Without Attack Labels
A new misbehavior detection scheme called PAMPOS uses a causal transformer-decoder trained solely on benign vehicle trajectories to identify falsification attacks in V2X networks. Unlike existing supervised methods, PAMPOS requires no labeled attack data, instead detecting anomalies via a top-K normalized scoring mechanism that pinpoints falsified kinematic features. Evaluated on all 19 attack types in the VeReMi++ dataset under rush-hour and afternoon scenarios, it achieves AUC up to 0.98 and F1-scores up to 0.95 for most attacks. The approach addresses the critical limitation of current learning-based MDSs that fail against unseen attacks.
Key facts
- PAMPOS is a causal transformer-decoder for misbehavior detection in V2X networks.
- It is trained only on benign VeReMi++ trajectories.
- No labeled attack data is needed for training.
- Anomaly detection uses a top-K normalized scoring mechanism.
- Evaluated on all 19 attack types in VeReMi++.
- Tested under rush-hour and afternoon scenarios.
- Achieves AUC up to 0.98 and F1-scores up to 0.95.
- Addresses failure of supervised MDSs against unseen attacks.
Entities
—