Maria Oikonomou, “Multilevel crash risk assessment under autonomous vehicle integration: From partial to full autonomy through urban traffic simulation”, PhD Thesis, School of Civil Engineering, National Technical University of Athens (November 2021 – January 2026).

The thesis develops a validated framework for multilevel crash-risk estimation during the transition from partial to full Autonomous Vehicle (AV) deployment, addressing the current absence of generalisable crash data for higher automation levels. The thesis begins with a systematic literature review of 54 studies, resulting in a detailed repository of AV behavioural parameters, including empirical, calibrated, theoretical and expert-based values. Building on this foundation, the thesis integrates traffic microsimulation, conflict-based Surrogate Safety Measures (SSMs), and a novel Time-To-Collision-based conflict-to-crash-risk conversion procedure. Validation is conducted using k-means clustering to compare simulated and observed crash patterns, demonstrating strong correspondence and confirming the robustness of the conversion method.

The validated framework is then applied to a high-fidelity microsimulation of the Athens city center, constructed from OpenStreetMap geometry and calibrated with detector flows, fleet composition data, pedestrian counts and public transport operations, and validated against Google Maps API travel times. Fifteen AV deployment scenarios are evaluated across varying Market Penetration Rates (MPRs) and SAE automation levels. Safety impacts are assessed through a multiscale methodology: (i) a road-level eXtreme Gradient Boosting (XGBoost) model with SHAP analysis quantifies the influence of traffic, infrastructure and automation on crash risk, and (ii) a conflict-level spatial analysis combining the Getis-Ord Gi* statistic and a binomial Generalised Additive Model (GAM) identifies crash hotspots. Results indicate nonlinear safety effects, with elevated crash risk during early and intermediate mixed-traffic stages due to behavioural heterogeneity, followed by substantial reductions once automated behaviour dominates and stabilizes network dynamics.

Overall, the thesis provides a validated, transferable methodological framework capable of supporting researchers, planners and policymakers in anticipating safety outcomes across evolving automation stages and informing the development of risk-aware deployment strategies for autonomous urban mobility.