Sensitivity of Sequence Models to Temporal Jitter in Automated Driving System Classification
A new arXiv preprint evaluates GRU, LSTM, and Transformer encoder models for classifying Level 2 automated driving systems from vehicle telematics data. While all models achieve high accuracy on clean data, the study finds that introducing temporal jitter—a type of realistic data corruption—causes a dramatic drop in performance across all models. The authors also present a modular framework for systematically testing model robustness to various telematics degradations.
Why it matters: The findings reveal a significant vulnerability in current AI-based monitoring systems for automated driving, with implications for safety and reliability in real-world deployments.
Full story at: arXiv Machine Learning ↗