Adaptive Physics-Informed Trajectory Reconstruction for Inferring Driver Aggressiveness (Apitra-Da) From Noisy Vehicle Kinematic Data

Abstract

Road traffic crashes claim over 1.3 million lives annually, with human factors responsible for more than 90% of incidents. Despite the growing availability of vehicle trajectory data enabling microscopic analysis of driving behavior, methodological challenges persist due to severe noise and missing vehicle specifications. This study introduces APITRA‑DA, a two‑stage adaptive physics‑informed framework designed to reconstruct noisy trajectories and infer driver aggressiveness without requiring vehicle specifications or behavioral labels. Applied to two expressway datasets where 97.6% and 59.9% of trajectories exhibited quality issues, the framework achieved 99% physics consistency, transforming degraded data into behaviorally interpretable trajectories. Noise reduction employed Butterworth filtering to preserve behavioral signals, while reconstruction integrated machine learning‑based vehicle capability estimation (R² > 0.78), physics‑based constraints, and driver‑adaptive Kalman filtering converging within three iterations. A novel aggressiveness parameter (α) quantifies how intensively drivers exploit vehicle capabilities, normalized for cross‑vehicle comparison. Validation demonstrated significant correlations with following distance (ρ = -0.45) and speed variance (ρ = +0.43), confirming α as a genuine behavioral metric. Results reveal drivers typically operate at ~60% of vehicle capacity, reserving unused capability for emergencies. Aggressiveness emerges as a dynamic, context‑dependent strategy—higher during acceleration than braking, decreasing with speed, and elevated in free‑flow conditions. Following distance and speed variance patterns provide key behavioral signatures, with aggressive driver proportions varying from 0.2% to 9.9% across datasets. APITRA‑DA enables applications in insurance telematics, traffic safety, emissions modeling, autonomous vehicle development, and simulation, advancing Vision Zero strategies by unlocking behavioral analytics from naturalistic trajectory data.