Semiautonomous Vehicle Risk Analysis: A Telematics-Based Anomaly Detection Approach

Area of Science:

• Transportation Engineering
• Data Science
• Insurance Risk Assessment

Background:

• Semiautonomous vehicles (SAVs) promise reduced accidents by minimizing human error.
• The shift to autonomous driving introduces new technological risks, challenging traditional insurance models.
• Historical data is insufficient for assessing the evolving risk landscape of SAVs.

Purpose of the Study:

• To investigate the unique risk structure of semiautonomous vehicles.
• To develop and apply a telematics-based anomaly detection model for assessing split risk profiles.
• To enable proactive risk reassessment by insurers using sensor-generated data.

Main Methods:

• Employed an unsupervised multivariate Gaussian (MVG) based anomaly detection model.
• Utilized accelerometer and GPS sensor data from manually driven vehicles to identify abnormal driving patterns.
• Inferred parameters for SAVs to determine split risk profiles, including location-based analysis.

Main Results:

• The MVG approach successfully quantifies vehicle risks based on the frequency and severity of anomalies.
• A split risk profile was determined, differentiating between human and technological error contributions.
• Location-based risk analysis provided a more comprehensive assessment of SAV risks.

Conclusions:

• The proposed telematics-based anomaly detection method addresses the challenge of quantifying SAV risks.
• This approach allows for the quantification of risks by analyzing observed anomalies and their severity.
• Findings support insurers in proactively reassessing risks associated with both human drivers and autonomous systems.