Explore the full website through grounded summaries and direct links to its sections, projects and research outputs.
PhD Researcher · Automotive Engineering · Technische Universität Berlin (TU Berlin)
Turning mobility data and simulation into safer decisions.
AV/CAV Safety · SUMO Simulation · Mobility Analytics · Explainable AI
This app turns the simulation results from my PhD at TU Berlin into an interactive experience. Whether you work with mobility data every day or are simply curious, you can explore the evidence at your own level.
Inside MSI
A short sequence of the maps and network tools used to explore the research.
SUMO, made easier
Select a packaged network or crop a real street area on the map, configure a scenario, and turn a live SUMO run into readable mobility, safety, and emissions evidence.
Who I Am
I’m Amirhossein Taheri, a PhD researcher in Automotive Engineering at Technische Universität Berlin (TU Berlin). My research explores how connected and automated vehicles can shape safer and smarter mobility systems.
I combine SUMO microsimulation, surrogate-safety analysis, mobility analytics, and AI-supported methods to transform complex mobility data into transparent, decision-relevant evidence.
Research architecture
My PhD centres on traffic simulation. Alongside it, I pursue traffic safety as an independent research line—connecting simulated systems with crash data, statistical inference, machine learning, and visual observation.
My PhD research platform uses microscopic traffic simulation to examine connected and automated vehicles in mixed traffic. It connects surrogate-safety analysis with headway and fleet scenarios, spatial hotspots, machine learning, SHAP interpretation, emissions, heat, and network-level mobility evidence.
Explore AV/CAV safety & simulation outputs ↓Across my papers I examine crash severity and duration, injury outcomes, vehicle effects, young-driver and rural-road risk, road-level heterogeneity, and spatiotemporal safety patterns.
Explore crash & road safety outputs ↓Connecting traffic flow, delay, vehicle composition, and energy use to environmental performance.
Explore urban climate & emissions ↓Exploring how mobility, congestion, waste heat, and exposure interact across urban space.
Explore urban climate & emissions ↓Turning simulation, crash, spatial, and visual data into interpretable patterns and decision-relevant evidence.
Explore mobility analytics & AI ↓Research outputs
Switch between peer-reviewed evidence and the projects that connect the methods, applications, and wider research questions.
The same categories are used in the research architecture, filters, and research guide.European Transport Research Review, 18(1), Article 9
International Journal of Intelligent Transportation Systems Research, 1-19
Transportation Letters · Published online 26 August 2026
Applied Sciences, 15(9), Article 4793
International Journal of Crashworthiness, 29(2), 198-209
Transportation Research Part F: Traffic Psychology and Behaviour, 97, 231-245
International Journal of Transport and Vehicle Engineering, 17(2), 102-110
Amirkabir Journal of Civil Engineering, 54(11), 4239-4252
Discrete Dynamics in Nature and Society, 2022(1), Article 2828277
Amirkabir University of Technology
Injury Prevention (BMJ) · Manuscript ip-2026-046183
Injury Prevention (BMJ) · Manuscript ip-2026-046270
Accident Analysis & Prevention · Manuscript AAAP-D-26-01122
Berlin SUMO–PALM case study · Traffic heat and urban microclimate coupling
arXiv 2508.09941
Transportation Research Board 2026 Annual Meeting
Transportation Research Board 2026 Annual Meeting · Tehran case study
12th International Conference on Urban Climate · Rotterdam
An interactive research product that turns prepared SUMO and SSAM outputs into accessible evidence for technical and non-technical users. It combines scenario and headway comparisons, TTC-based surrogate-safety indicators, standardized Gaussian KDE hotspot ranking, geographic network mapping, local and whole-network 3D lenses, and precomputed LightGBM–SHAP interpretation.
Simulation records are harmonized across fleet scenarios and desired headways, summarized by interaction type and safety threshold, transformed from SUMO coordinates for spatial exploration, and grouped into ranked conflict concentrations. Offline models compare microscopic, policy-lever, and combined feature sets; the application loads the validated prepared outputs without retraining models for each visitor.
This ongoing research programme investigates how autonomous vehicles can adjust their following headway according to traffic conditions and location-specific risks. Using the Berlin-Charlottenburg SUMO network, it progresses from local hotspot intervention to spillover analysis and whole-network optimization.
Tests whether increasing AV headway near selected high-risk locations can reduce near-miss risk without substantially disrupting traffic.
Examines whether local safety improvements transfer conflicts, queues or disturbances to surrounding and downstream areas.
Extends adaptive headway across the Charlottenburg network to balance safety, mobility and model-based emissions while limiting unintended impacts elsewhere.
SUMO microsimulation, ACC-based autonomous vehicles, TraCI adaptive control, mixed traffic, surrogate-safety indicators, spatial risk analysis, traffic-performance assessment, emissions modelling and multi-objective optimization.
A map-first interface for running mobility experiments without handling SUMO configuration files directly. Users can select a packaged network or draw a real OpenStreetMap area, adjust demand, automated-vehicle share, following time and speed, then run SUMO and compare travel time, waiting, CO₂, speed and TTC-based interactions with a fixed reference.
The selected map crop is downloaded from OpenStreetMap and converted into a SUMO network. The interface translates the approved controls into route and vehicle definitions, executes SUMO locally, parses trip, emissions and surrogate-safety XML outputs, and presents a reference comparison. A separate robustness lab uses matched random seeds and paired uncertainty; the optional EZ Agent only helps prepare a constrained study plan and never runs without approval.
An interdisciplinary collaboration coupling SUMO traffic dynamics, fuel use and vehicle-level emissions with PALM’s building-resolving climate grid. Traffic outputs are converted into sensible heat and moisture emissions to evaluate localized effects on near-surface temperature and urban thermal conditions in Berlin.
The Berlin case study couples vehicle-level fuel use with PALM’s urban-climate grid. District-scale daily mean warming is small under the investigated conditions, while localized responses are stronger along traffic corridors and near intersections, especially during weak-wind periods.
Developed and evaluated an LRT-CNN-VD framework using Pennsylvania crash-sequence data. The Bayesian 1D-CNN combines the local reparameterization trick, variational inference and variational dropout for accurate, uncertainty-aware crash-clearance duration modelling.
The model was compared with CNN, CNN with dropout, Random Forest Regression and KNN Regression. It achieved the best test RMSE and MAPE with substantially lower tuning and training time than dropout regularization; SHAP identified influential factors including temperature, humidity, visibility, crash severity, weather and incident distance.
Originating in a master’s thesis I supervised and later developed into a co-authored publication, this study examined injury severity among drivers aged 17–24 using England’s STATS19 data from April 2019 to February 2022.
Standard CART was compared with Random Undersampling of the Majority Class CART to address severe class imbalance. RUMC-CART improved identification of serious injuries; important predictors included casualties, vehicles involved, vehicle type, speed limit, lighting and urban–rural context.
Investigated how vehicle type and weight influence driver injury severity in fixed-object, rollover and two-vehicle crashes. Six years of police-reported rural crash data were used to develop a logistic-regression crashworthiness index across vehicle categories.
Crashworthiness varied substantially by crash scenario. SUVs and trucks generally offered stronger occupant protection, while pickup trucks performed less well in several comparisons; heavier vehicles were not uniformly safer, with the 1,500–2,000 kg group showing the lowest estimated injury/death odds in fixed-object crashes.
Developed a hybrid forecasting framework that combines LSTM neural networks with ARIMA residual modelling using 15-minute congestion-index observations from a large metropolitan road network.
The approach was designed to capture nonlinear temporal patterns and the remaining time-series structure, and was evaluated against standalone LSTM forecasting using MAE, RMSE and R². The project contributes to traffic-state forecasting and hybrid machine learning for mobility decision support.
Academic service
I contribute to scholarly quality through peer review in transportation, traffic safety, mobility analytics, and data-driven research, including work associated with Taylor & Francis, Springer Nature, and transportation conferences.
Research on mobility systems, behaviour, operations, and future transport.
Crash outcomes, surrogate safety, risk modelling, and safety performance.
Statistical, machine-learning, spatial, and simulation-based evidence.
Academic development
I support emerging researchers in study design, simulation and data analysis, interpretation, and academic writing. The selected work below spans automated mobility, crash severity, machine learning, and policy evaluation.
Assessment of the Safety Impacts of Autonomous Vehicles in Relation to Market Penetration and Vehicle Fleet Composition: A SUMO Scenario Study for Berlin-Charlottenburg
Evaluation and Sensitivity Analysis of Calibrated Traffic-Safety Parameters in Connected and Autonomous Vehicle Simulations
COVID-19 Policy Measures and Road Traffic Injury Severity in Great Britain: A Hybrid Logistic Regression Approach
The Impact of Connected and Automated Vehicles on Road Safety: A Meta-Analysis Considering Market Penetration Rate
Machine Learning for Enhancing Road Safety: Evaluating Crash Severity Prediction Models - A Case Study of Australia
Analysis of Imbalanced Traffic Data: A Comparative Study of Machine Learning Models for Predicting Accident Severity in England
Open channel
For research conversations, collaborations, simulation work, or questions about a paper, use the email button below to contact me at Technische Universität Berlin.
