Pelella, Fabrizio (2025) Towards a skilled and attentive human driver model when interacting with the infrastructure: longitudinal acceleration models with behavioural adaptation to horizontal road geometry. [Tesi di dottorato]

Documento PDF pelella_fabrizio_38.pdf Visibile a [TBR] Amministratori dell'archivio Download (5MB) | Richiedi una copia

Abstract

In a transportation context moving towards an increasingly connected, cooperative, and automated mobility, the definition of the skilled and attentive human driver when interacting with the infrastructure is essential for road design, road safety studies, and traffic simulation that will concern a forthcoming mixed scenario where both automated and traditional vehicles will coexist. To this end, we integrated the effects of horizontal road geometry into longitudinal acceleration models (i.e., car-following models) by developing a parsimonious desired speed layer as a function of the radius of curvature. A spatial anticipation mechanism and an indifference radius threshold are also included to describe realistically the adaptive driving behavior when approaching or exiting from curves. We integrated the desired speed layer into the Intelligent Driver Model (IDM) and its modification (MIDM), the Gipps’ model, the Optimal Velocity Model (OVM), and the Full Velocity Difference Model (FVDM). We assessed the accuracy of the curvature-augmented models (namely, IDM-r, MIDM-r, Gipps-r, OVM-r, and FVDM-r), using a literature-established calibration framework based on the Least Squares Error method. This research focuses on a motorway setting and considers two experimental contexts: (i) an isolated vehicle scenario, in which a single vehicle interacts exclusively with the infrastructure, and (ii) a leader-follower scenario, including separately free-flow and car-following dynamics. In this framework, free-flow refers to leader-follower pairs with minimal interaction, whereas car-following describes conditions in which the leader has a significant influence on the ego-vehicle’s behaviour. Both isolated vehicle and leader-follower contexts were addressed using corresponding experimental data: we used single vehicle trajectories collected from the Asse Mediano for the former, while we used leader-follower pairs provided by the ZEN Traffic Data project for the latter. We further classify the leader-follower pairs into free-flow and car-following trajectories according to specific selection criteria. The results show that, in isolated vehicle conditions, the improved formulations outperform their basic counterparts for all the trajectories, being able to reproduce excellently the ego vehicle’s motion. Regarding free-flow and car-following dynamics, the improvement is evident, even though less pronounced than in the isolated vehicle context. Indeed, the quality of data plays a non-negligible role in determining lower calibration errors of the curvature-augmented models rather than their basic counterparts. Nevertheless, the calibration results in both free-flow and car-following regimes are more accurate when considering settings with a lower radius of curvature. Overall, the findings confirm that the explicit inclusion of horizontal curvature allows for a more realistic representation of human driving behaviours, providing a useful basis for traffic modelling and simulation, as well as for road safety assessment that involves both traditional and automated vehicles when interacting with road geometry.

Downloads

Actions (login required)

Modifica documento