Digital Twin for Synthetic Data Generation

This paper presents a digital twin framework for synthetic data generation aimed at accelerating the development and validation of automated driving systems.

The work investigates how virtual environments and physics-based simulation can support safety-critical testing of sensing and perception systems, particularly in scenarios involving vulnerable road users and varying environmental conditions.

Using Simcenter Prescan and synthetic KITTI-style datasets, the research evaluates the robustness of camera-based perception pipelines under both parametric and aleatoric uncertainty.

Experimental results demonstrate that weather and illumination variations introduce significantly greater uncertainty than differences between vulnerable road user categories, highlighting the importance of synthetic data for robustness assessment in autonomous vehicle perception systems.