National Technical University of Athens (NTUA)
Multilevel crash risk assessment under autonomous vehicle integration: from partial to full autonomy through urban traffic simulation
Abstract
dc:descriptionAutonomous Vehicles (AVs) are expected to transform urban mobility by reducing human-error-related crashes and improving traffic efficiency. However, the absence of generalizable crash records for higher levels of automation limits the ability to evaluate their safety using traditional, data-dependent methods. To address this challenge, this thesis develops and validates a multilevel, simulation-based framework for estimating the crash risk of AV deployment across the transition from partial to full autonomy. A systematic literature review was conducted to establish a comprehensive repository of AV behavioural parameters, derived from 54 studies. Beyond numerical parameter ranges, the repository integrates information on parameter sources (empirical, calibrated, theoretical and expert-based), simulation software, driving-behaviour models, network context, traffic demand, SAE level and Market Penetration Rate (MPR) effects, providing an evidence-based foundation for modelling AV behaviour in traffic microsimulations. The methodology developed in this thesis comprises: (i) traffic microsimulation, (ii) conflict-based surrogate safety measures, (iii) a novel Time-To-Collision (TTC) event-level crash-risk conversion procedure and (iv) external validation using k-means clustering to compare simulated and observed crash patterns. The validation confirms strong agreement between simulated and field crash-risk distributions, establishing the robustness of the crash-risk conversion method. Following validation, the methodology is applied to assess AV impacts using a high-fidelity microsimulation model of the Athens city centre, constructed from OpenStreetMap geometry and calibrated using traffic detector flows, field fleet composition measurements, pedestrian counts, public transport operations and validated through Google Maps travel times. Fifteen AV deployment scenarios were simulated, covering different MPRs of SAE 0-1, 2-3, 4-5, and 4-5 with an aggressive operational profile. Crash risk was then evaluated within a multi-scale safety-assessment framework, including: (i) road-level, using an eXtreme Gradient Boosting (XGBoost) model with SHapley Additive exPlanations (SHAP) to quantify the influence of traffic, geometric and automation variables, and (ii) conflict-level, using spatial crash hotspot analysis using Getis-Ord Gi* statistic and a binomial Generalized Additive Model (GAM) to detect localized high-risk clusters. Results show that safety does not improve monotonically with increasing automation. Instead, crash risk follows a nonlinear trajectory: it increases during the early and intermediate stages of mixed-traffic conditions due to behavioural heterogeneity, and decreases substantially only at high AV penetration, when automated behaviour becomes dominant and stabilises network interactions. These findings suggest that automation primarily acts as a system stabiliser, reducing variability and weakening the spatial concentration of crash hotspots, rather than eliminating crash risk immediately. Overall, this thesis delivers a validated and transferable framework for multilevel safety assessment of autonomous urban traffic. The approach could support researchers, planners and policymakers in anticipating safety outcomes across automation stages, informing the design of risk-aware deployment strategies and contributing to a safer transition toward autonomous mobility.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Oikonomou, Maria
- Οικονόμου, Μαρία
Subjects
dc:subject × 22- Αυτόνομα οχήματα
- Οδική ασφάλεια
- Προσομοίωση κυκλοφορίας
- Εναλλακτικοί δείκτες ασφαλείας
- Χρόνος έως την εμπλοκή
- Εκτίμηση κινδύνου ατυχήματος
- Μηχανική μάθηση
- Χωρική ανάλυση σημείων συγκέντρωσης ατυχημάτων
- Autonomous vehicles
- Road safety
- Traffic simulation
- Surrogate safety measures
- Time-to-collision
- Crash risk modelling
- Machine learning
- Spatial hotspot analysis
- Επιστήμη Πολιτικού Μηχανικού
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη και Τεχνολογία των μεταφορών
- Civil Engineering
- Engineering and Technology
- Transportation science and Technology
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60972