Twin the Drive: An HGBR-based Machine Learning Model as a Vehicular Digital Twin for Risk Detection
Cansu Demir; Narges Mehran; Alexander Meschtscherjakov (2025): Twin the Drive: An HGBR-based Machine Learning Model as a Vehicular Digital Twin for Risk Detection In: Adjunct Proceedings of the 17th International Conference on Automotive User Interfaces and Interactive Vehicular Applications.
Vehicular Digital Twins (VDTs) can enhance safety in automated driving by modeling and analyzing driving behavior in real time. However, most current approaches rely on complex machine learning (ML) models that are difficult to interpret and deploy in human-centered interfaces. In this paper, we present a proof-of-concept VDT that uses a simple, interpretable ML model from the category of boosting to detect three key risk indicators: unsafe time headway, harsh braking, and time-to-collision (TTC) for both consecutive vehicles. Our approach is motivated by the need for real-time risk detection in cooperative in-vehicle intelligent agents (IVIAs). We evaluate the model using the NGSIM dataset and show that it closely approximates unsafe time headway and braking events, with a conservative prediction bias that may be beneficial in risky contexts. TTC risks are under-predicted, reflecting limitations in modeling complex interactions with simple features. This research demonstrates the feasibility of interpretable VDTs for near real-time behavioral risk detection, laying the groundwork for future integration with IVIAs.
