Coexisting Microvehicles and Pedestrians: A Novel Modeling Approach

Microvehicles, such as e-scooters and bicycles, are increasingly used in urban transport systems. Existing models were primarily developed either for pedestrians-only or motor vehicles-only and cannot fully capture the dynamics of microvehicles, particularly when they operate in shared spaces also used by pedestrians. This paper proposes a hybrid microscopic modeling framework for microvehicles’ and pedestrians’ trajectories based on a combination of physical interaction principles and deep learning techniques. A Long Short-Term Memory (LSTM) model is developed using space discretization and interaction variables to predict user trajectories in shared environments. The proposed framework was evaluated using observations collected through a semi-controlled experiment conducted at the University of Patras. Nearly 300 participants generated 553 trajectories involving pedestrians, bicycles, and e-scooters that were used for model calibration and validation. The model achieved satisfactory trajectory prediction accuracy with an RMSE below 0.20 m in testing scenarios. The model was compared to an adapted Social Force Model (SFM) for microvehicle and pedestrian interactions and showed better performance. The findings highlight important behavioral differences between pedestrians, bicycles, and e-scooters in terms of interaction distances and fields of view. The proposed framework can support future applications in infrastructure design, safety assessment, and traffic management.