SLM-Based Framework Detects GNSS Spoofing Attacks in Autonomous Vehicles
A recent study published on arXiv (arXiv:2608.17092) introduces a framework for a small language model (SLM) designed to identify GNSS spoofing attacks in autonomous vehicles (AVs). The research tackles the vulnerabilities of AVs by utilizing independently obtained driving states from multiple sensors in conjunction with GNSS data, which are transformed into structured semantic narratives to facilitate attack detection. This framework was evaluated against large language models (LLMs) across five categories of attacks and was tested with data from unfamiliar geographical locations, suggesting a degree of generalizability. The SLM approach prioritizes computational efficiency, aiming to bolster the resilience of AV positioning systems against spoofing, and underscores the convergence of AI, automotive engineering, and cybersecurity.
Key facts
- The study proposes an SLM-based framework for GNSS spoofing detection in autonomous vehicles.
- It compares vehicle behaviors from GNSS and other sensing sources.
- Driving states are converted into structured semantic narratives for the SLM.
- Performance is benchmarked against fine-tuned LLMs on identical training data.
- Five attack classes are considered: no attack, overshoot attack, stopped attack, turn-by-turn attack, and wrong-turn attack.
- The framework was evaluated with geographically unseen field data.
- The research is documented as arXiv:2608.17092.
- The approach aims to be more efficient by using small language models.
Entities
—