GRU, LSTM, and Transformer Models Achieve High Accuracy in Classifying Automated Driving Systems
A recent study on arXiv investigates the efficacy of three machine learning models—Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), and Transformer encoders—in classifying Level 2 automated driving systems using telematics data. The analysis includes three commercial platforms: Comma.ai's Openpilot, Tesla Autopilot, and Cadillac Super Cruise, as well as manual driving for comparison. The models demonstrated impressive performance with macro F1-scores of 0.92 for GRU, 0.90 for LSTM, and 0.93 for the Transformer. This research highlights the critical need for independent oversight in the rapidly evolving field of automated driving technologies to ensure safety.
Key facts
- The paper evaluates GRU, LSTM, and Transformer encoder models for classifying automated driving systems.
- The systems classified are Comma Openpilot, Tesla Autopilot, Cadillac Super Cruise, and manual driving.
- Macro F1-scores on clean data: 0.92 (GRU), 0.90 (LSTM), 0.93 (Transformer encoder model).
- The study uses vehicle telematics data alone for classification.
- The research is motivated by the need for independent monitoring of ADSs for safety, compliance, insurance, and anomaly detection.
- The paper is available on arXiv with ID 2607.28665.
- The announcement type is cross.
- Future Software Defined Vehicles (SDVs) may run multiple ADSs, both native and aftermarket.
Entities
Institutions
- arXiv
- Comma.ai
- Tesla
- Cadillac