Elevation Aware 2D/3D Co-simulation Framework for Large-scale Traffic Flow and High-fidelity Vehicle Dynamics

Reliable testing of autonomous driving systems requires simulation environments that combine large-scale traffic modeling with realistic 3D perception and terrain. Existing tools rarely capture real-world elevation, limiting their usefulness in cities with complex topography. This paper presents an automated, elevation-aware co-simulation framework that integrates SUMO with CARLA using a pipeline that fuses OpenStreetMap road networks and USGS elevation data into physically consistent 3D environments. The system generates smooth elevation profiles, validates geometric accuracy, and enables synchronized 2D-3D simulation across platforms. Demonstrations on multiple regions of San Francisco show the framework’s scalability and ability to reproduce steep and irregular terrain. The result is a practical foundation for high-fidelity autonomous vehicle testing in realistic, elevation-rich urban settings.

💡 Research Summary

The paper introduces an automated, elevation‑aware co‑simulation framework that tightly integrates the microscopic traffic simulator SUMO with the high‑fidelity 3D environment simulator CARLA. The authors address a critical gap in autonomous‑vehicle (AV) testing: most existing simulation platforms assume flat terrain, which fails to capture the challenges posed by cities with significant topography such as San Francisco.

The core of the framework is a multi‑stage pipeline that fuses openly available geographic data sources. First, road network geometry is extracted from OpenStreetMap (OSM) in UTM coordinates. Second, a Digital Elevation Model (DEM) from the USGS is queried for each road point. Because DEMs are raster grids, the authors employ bilinear interpolation to obtain a smooth elevation value for any (x, y) location. The resulting 3‑D point cloud (x, y, z) is constructed by a simple stacking operation (Equation 1) and stored as a road‑segment list.

To ensure that the generated road profiles are physically plausible, the pipeline enforces gradient limits derived from civil‑engineering standards (8 % for highways, 12 % for arterials, 15 % for residential streets). When a segment exceeds these limits, a three‑point moving‑average smoothing (Equation 9) is applied iteratively until all gradients satisfy the constraints. This step eliminates noise inherent in raw DEM data while preserving the overall terrain shape.

The 3‑D road network is then imported into MathWorks’ RoadRunner, which automatically creates CARLA‑compatible meshes, lane markings, traffic signs, and other infrastructure elements. Because RoadRunner can output directly to CARLA’s asset format, the authors avoid manual modeling and achieve a fully procedural environment generation.

The co‑simulation architecture synchronizes SUMO and CARLA via ROS 2 messages. SUMO runs the background traffic for thousands of vehicles in 2‑D, while CARLA simulates the ego vehicle with photorealistic sensors (LiDAR, RGB cameras, radar, GPS). Time stamps and vehicle states are exchanged at a configurable frequency, ensuring that the ego vehicle perceives realistic traffic participants that obey the same road geometry and elevation.

Validation is performed on several districts of San Francisco, a city known for steep hills and irregular road grades. The authors compare the generated elevation profiles against the original DEM using Mean Absolute Error (MAE ≈ 0.27 m), Root Mean Square Error (RMSE ≈ 0.34 m), and maximum error (≈ 1.2 m). Gradient compliance is reported at 98 % of road segments, confirming that the smoothing step successfully respects design limits.

Beyond geometric fidelity, the paper demonstrates the impact of elevation on perception. LiDAR point clouds exhibit occlusions on hill crests, and camera fields‑of‑view are distorted by steep slopes. When these elevation‑aware scenes are used to test perception pipelines, object‑detection recall improves by roughly 12 % compared with flat‑world baselines, and vehicle control signals (speed, acceleration) align more closely with real‑world measurements on sloped roads.

The contributions are threefold: (1) a semi‑automated, modular pipeline that converts OSM + DEM data into CARLA‑ready 3‑D environments; (2) a scalable SUMO‑CARLA co‑simulation that can handle city‑scale traffic while preserving high‑fidelity sensor realism; and (3) an evaluation showing that elevation‑aware simulation yields more accurate perception and control results. The framework is open‑source, extensible to any region with publicly available GIS data, and positions itself as a practical foundation for rigorous AV testing in elevation‑rich urban settings.