TL;DR: Most scenario generators for autonomous-vehicle testing change the traffic and keep the road fixed. I generate the road itself. Topo-DiffRoad produces realistic highway mainlines, ramps and interchanges with controllable risk geometry (curvature, grade, vertical curves) and exports them straight to CARLA. In closed-loop tests, these roads caused far more tracking errors and lane departures than existing test maps.

Motivation

Simulation-based testing of autonomous vehicles (AVs) mostly varies the dynamic parts of a scene: cut-ins, pedestrians, aggressive drivers. The road usually stays a fixed background. On highways, though, the road itself is a major risk factor. At high speed and with long braking distances, sharp horizontal curves, steep grades and short vertical curves amplify small control errors into lane departures and crashes.

Existing road generators mostly target urban intersections and cannot control these alignment properties directly. That leaves a gap: there was no way to generate realistic highway roads that are risky by design, which is exactly what is needed to stress-test AV control stacks.

Approach

The work closes the loop from crash data to test scenarios: identify which road features drive crashes, generate roads with those features, and verify in simulation that the generated roads really are harder to drive.

1. Real highway data

I built a dataset of 1,937 highway scenes from OpenStreetMap covering the full freeway networks of Washington, Oregon and California. It has three scene types: 1,228 mainline segments, 382 ramp junctions and 327 complex interchanges. Interchanges were located by clustering motorway-junction nodes with DBSCAN. Each scene was cropped, projected to UTM and resampled to fixed-length centerlines.

2. Topo-DiffRoad: a two-stage conditional diffusion model

Generating a multi-road interchange in a single diffusion pass tends to produce misaligned endpoints and broken connections. I therefore split generation into topology and geometry:

• Stage 1 (SkeletonNet, Transformer): generates the endpoint skeleton, i.e. the start and end points of every road in the scene. This fixes how roads connect, run in parallel or cross.
• Stage 2 (RoadGenNet, 1-D U-Net): generates each road’s full centerline, conditioned on the skeleton, the scene type and a target maximum curvature. The network uses per-road endpoint embeddings, bottleneck self-attention, FreeU re-weighting on skip connections, and a smoothness regularizer on the predicted geometry.

Both stages use DDPM with v-prediction and a cosine noise schedule.

3. From risk factors to a drivable 3D scene

Risk factors identified from crash-frequency modelling (see the related project below) enter the pipeline at three levels:

The output is a standard .xodr file that loads directly into CARLA or SUMO, with elevations aligned at every junction.

4. Closed-loop testing in CARLA

To isolate the effect of road geometry, I ran single-vehicle closed-loop tests: no other traffic, a fixed vehicle (Tesla Model 3) at 72 km/h, and two lateral controllers, LQR (optimal control) and Stanley (geometric feedback). Generated scenes were compared against DiffRoad (a state-of-the-art road generator), CARLA Town04/06 and CommonRoad real-road scenarios.

Results

Realistic, topologically valid roads

Topo-DiffRoad matches the real distributions of road length, point spacing and curvature most closely. Curvature JSD drops by 44% compared with DiffRoad, and centerlines are about 15× smoother. Hausdorff distance is on par with DiffRoad.

Harder for controllers to drive

• Generated scenes cover risk geometry that existing maps rarely contain: most reach grades of 4–5.5%, and maximum curvature ranges from 2.5 to 25.7°/100 ft.
• Mean lateral tracking error was ~3.5× (LQR) and ~4.5× (Stanley) that on DiffRoad scenes, and higher still than on CARLA Town and CommonRoad.
• Road geometry translates directly into control difficulty: heading change rate correlates with mean lateral error at Spearman ρ = 0.97 (LQR) / 0.94 (Stanley).

More safety-critical events

My role

Sole author of the thesis. I designed and built the OSM data pipeline, the Topo-DiffRoad model, the rule-based 3D and OpenDRIVE generator, and the CARLA evaluation framework and metrics.

Limitations and next steps

• Region transfer: risk thresholds were calibrated on U.S. crash and map data; applying the method to other road networks (e.g. China) needs local calibration.
• Beyond geometry: the scenes contain the road only. Adding guardrails, signs, vegetation and occlusion would allow testing perception and localization, e.g. sight-distance-limited curves.
• Richer testing: combining the risky road networks with dynamic traffic, and evaluating end-to-end driving stacks rather than only lateral controllers.

Related project

The risk factors and thresholds used here come from my crash-frequency modelling work: Traffic Crash Data Augmentation with Multi-type Variables using Hybrid VAE-Diffusion Generative Neural Networks for Enhancing Crash Frequency Modeling, Accident Analysis and Prevention, 2026 (co-first author). It used SHAP on the augmented crash model to identify curvature, grade and vertical-curve length as key risk factors for high-crash highway segments.

TRB 2026 poster. Click to open the full-resolution version.

At a glance

• Question: Are walking and cycling amenities linked to diabetes, and through what pathway?
• Data: 32 local government areas (LGAs) in Victoria, Australia, combining OpenStreetMap, the VISTA travel survey, and PHIDU health data
• Method: Partial least squares structural equation modeling (PLS-SEM) with mediation analysis
• Finding: No direct effect. Amenities matter for health through behaviour: more active travel → lower health risks → lower diabetes prevalence.
• Presented: Poster session, TRB 105th Annual Meeting, Washington, D.C., January 2026

Motivation

Type 2 diabetes is strongly tied to physical inactivity, and active transport is one of the few ways cities can build physical activity into everyday life. Many studies link the built environment to physical activity, or to health outcomes, but few model the full chain between them empirically. Existing pathway frameworks (e.g., Frank et al., 2019) remain largely conceptual.

This study tests that chain quantitatively, and it uses objective amenity measures instead of the self-reported perceptions common in prior work.

Approach

Building on the behavioural pathway in Frank et al. (2019), I modeled four linked constructs:

I chose PLS-SEM over covariance-based SEM because it handles small samples and non-normal data. Both apply at the LGA level (n = 32). Models were estimated in SmartPLS 4. A second specification controls for the share of elderly residents to check for confounding by age.

Key Findings

• Amenities → active travel. LGAs with denser walking and cycling amenities show substantially more active travel (β = 0.63, p < .001).
• Active travel → health risks. More active travel is associated with lower obesity and psychological distress (β = −0.27, p = .012; −0.41 after controlling for age).
• Health risks → diabetes. Health risks strongly predict diabetes prevalence (β = 0.61, p < .001).
• Indirect-only mediation. The direct path from amenities to diabetes is not significant (β = −0.05, p = .84), but the sequential indirect path is (β = −0.107, 95% CI [−0.234, −0.017]).

Why it matters: Walking and cycling infrastructure is unlikely to improve health on its own. Its benefit depends on whether people actually use it. This supports the WHO’s call to invest in walking and cycling, and suggests planners should evaluate amenities by the behaviour they generate, not just by how much is built.

Limitations & Next Steps

This is an ecological, cross-sectional analysis of 32 areas, so the results describe area-level pathways rather than individual causal effects. Next steps I’m interested in: individual-level and longitudinal data, and additional mediators such as perceived safety, social cohesion, and accessibility.

My Role

First author. I conceptualized the study, developed the conceptual model and hypotheses, collected and processed the OpenStreetMap amenity data, specified and estimated the PLS-SEM models, and wrote the original draft. Co-authors supervised the work and contributed data curation (Y. Wang) and review and editing (both).

Crash frequency modeling is essential for identifying high-risk road segments and improving traffic safety. However, crash datasets are often plagued by excessive zero observations — most road segments record zero crashes — which undermines model accuracy.

Existing approaches (statistical models, resampling techniques) are limited by strict distributional assumptions or fail to handle the multi-type nature of crash data (count, ordinal, nominal, and real-valued variables). This project proposes a data-driven approach using TabSyn, a hybrid VAE-Diffusion generative neural network, to augment crash data while preserving inter-variable correlations.

Approach

• Transformed multi-type crash variables into a unified continuous latent space through a tokenizer-based embedding and VAE encoding.
• Applied a Diffusion model in the latent space to generate high-fidelity synthetic crash records.
• Integrated the augmented data with XGBoost for crash frequency prediction.
• Compared against state-of-the-art generative models (CTGAN, TVAE, GReaT, StaSy) and statistical baselines (ZIP, GAM-Poisson).

Key Findings

• TabSyn achieved the best distributional and structural fidelity among all benchmark generative models.
• The proposed framework achieved the lowest prediction error, particularly for Top-K high-risk segment identification.
• Results demonstrate that deep generative augmentation significantly outperforms traditional resampling and statistical methods for imbalanced crash data.

Presentation

Related Publications

Chen J*, He Q*, Liu P, Ma W, Zheng N, Liu P, Pu Z†. (2026) “Traffic Crash Data Augmentation with Multi-type Variables using Hybrid VAE-Diffusion Generative Neural Networks for Enhancing Crash Frequency Modeling.” Accident Analysis and Prevention, 231: 108508. Link

He Q, Chen J, Liu P, Ma W, Pu Z†. (2025) “Enhancing Crash Frequency Modeling with Mixed-Type Data Using a Hybrid VAE-Diffusion-Based Generative Approach.” Poster, TRB 104th Annual Meeting, Washington, D.C. (Paper No. TRBAM-25-06171)