Assessing Usage Spectrum and Regimes Severity in e-Scooters: a Learning-based Approach

E-scooters have recently spread as light vehicles for urban micro-mobility, gaining popularity due to their accessibility, ease of use, and eco-friendliness. However, concerns about safety and sustainability have also arisen. The increasing incidence of accidents and injuries, often linked to inexperienced and aggressive driving styles, underscores the need for active monitoring of driving behavior. Additionally, life cycle assessments reveal significant environmental impacts associated with e-scooter production and charging processes, emphasizing the importance of maximizing the vehicle’s lifespan to ensure its overall sustainability. To tackle these challenges, we propose an innovative real-time solution for automatically classifying the different e-scooter maneuvers and assessing the aggressiveness of the driving style. The maneuver recognition approach relies on machine learning techniques and leverages a multi-window approach to accommodate different dynamic time scales. On the other hand, severity assessment is based on specifically designed mathematical indicators, offering a multidimensional and interpretable perspective of the vehicle’s dynamic status. When applied to a real-case scenario involving trials with 31 volunteers and two distinct on-the-market e-scooters, our framework proves effective both in maneuver recognition and severity assessment.