Abstract
Automated vehicles (AVs) may improve urban road safety, yet aggressive calibration and mixed-traffic interactions may produce nonlinear patterns in TTC-defined surrogate events. This study examines how AV penetration, vehicle-size composition, and desired headway calibration shape simulated TTC-defined surrogate outcomes. Using SUMO microsimulation for Charlottenburg, Berlin, with TTC-based analysis, LightGBM modeling, and spatial hotspot mapping, we evaluate 12 scenarios varying AV penetration (0–100%) and the composition of small AV12 and larger AV46. Both AV classes share the same driving parameters and differ primarily in physical dimensions. The scenario matrix is evaluated at AV desired headways of τ = 0.6, 0.8, and 1.0 s, yielding 36 nominal scenario–headway combinations and 34 effective configurations; human-driven vehicle (HDV) headway remains fixed at 1.0 s. Under 0.6 s calibration, total TTC-defined conflict frequency declines with increasing AV penetration, reaching reductions of approximately 45% under full automation. However, mixed-traffic configurations at the 0.6 s headway produce a larger proportion of selected short-TTC events (TTC < 0.5 s). Sensitivity analysis using minTTC thresholds of 0.3, 0.5, and 0.75 s shows that the 0.6 s setting consistently produces the greatest concentration of short temporal margins, whereas 0.8 s represents an intermediate profile. Spatially, conflicts shift from network-wide dispersion under HDV-dominant conditions to localized concentration at critical junctions under AV-dominant traffic. Differences between AV12- and AV46-dominant configurations are comparatively small and vary across headway and TTC-threshold settings. LightGBM identifies relative speed and desired headway as the strongest predictive associations. The findings highlight the context- and calibration-dependent nature of simulated AV surrogate-safety outcomes, particularly the roles of market penetration, desired headway, and vehicle-size composition.
What the study examines
Penetration, vehicle size and headway.
A calibrated SUMO model of Berlin–Charlottenburg combines TTC-defined surrogate-event analysis, spatial hotspot mapping and LightGBM modelling. The study varies AV market penetration and the mix of smaller AV12 and larger AV46 vehicles, which share the same driving parameters and differ primarily in physical dimensions.
- At the 0.6 s AV headway, increasing AV penetration reduces total TTC-defined event frequency, while mixed traffic shows a greater proportion of selected short-TTC events.
- Headway sensitivity changes the concentration of short temporal margins. The 0.8 s setting represents an intermediate profile, rather than an established universal optimum.
- Conflict patterns become more concentrated at critical junctions under AV-dominant conditions. LightGBM identifies relative speed and desired headway as the strongest predictive associations.
These are simulation-based, calibration-dependent surrogate outcomes. They should not be read as measured reductions in real-world crashes or as causal effects inferred from feature importance.
Interactive study guide
Explore the headway findings
Change the illustrated fleet share and select a headway to read the corresponding study context.
Mixed traffic
0.8 s: an intermediate profile
The reported headway sensitivity places 0.8 s between the shorter- and longer-headway profiles. It does not establish a universally optimal headway.
The study compares different fleet compositions at each penetration level; no scenario-specific value is calculated by this illustration.
This guide summarizes reported patterns and illustrates fleet composition. It does not run SUMO or estimate conflict counts, TTC probabilities or crash risk. Use the paper and MSI for the underlying methods and prepared outputs.
The paper and the research product
Explore the evidence in MSI.
Mobility Safety Intelligence is my PhD research product for exploring the prepared simulation evidence associated with this study. It brings together scenario and headway comparisons, TTC thresholds, hotspot rankings and network views in an accessible interface.
The app also includes tools for exploring the prepared outputs, including Street Lens, 3D network views and offline LightGBM–SHAP interpretation. Cite the published paper for the study’s design, findings and limitations.
Source and access
Published article · CC BY 4.0
Abstract and study findings: Taheri, Dorn, Karimi, Budde, Milakis and Müller (2027), Accident Analysis & Prevention, 239, 108835. Version of record · Creative Commons Attribution 4.0. The interactive guide above is an authored summary. An earlier SSRN preprint remains available; the journal DOI is the reference for the published study.