Digital Twin-Based Simulated Automotive Radar for Virtual Testing

Abstract

Modern vehicles are increasingly characterized by software-based systems, and perception sensing technology plays an important part in the safety and the driving experience. The development of a reliable and efficient methodology for perception technology in a virtual environment holds the potential to accelerate the development of autonomous vehicles while ensuring safety and reliability. Between these perception systems, the development and validation of radar systems for autonomous driving is a major challenge due to the complex and dynamic nature of real-world driving scenarios. Within the scope of this thesis, a digital twin-based simulated environment for automotive radar perception analysis has been implemented. This work presents a method for the parameterization and validation of the digital twin of the selected test scenarios defined by Euro NCAP. The results are analyzed to evaluate the simulation-to-reality gap, indicating how closely the simulated data matches real-world observations. Additionally, the effects of sensor fidelity are examined to understand how accurately the sensor models replicate real-world performance. Lastly, a sensitivity analysis is conducted to identify critical elements. The outcome of the thesis indicates that virtual testing is a promising approach for radar sensor model analysis based on the idea of digital twin, especially using high-fidelity twins.