Virtual Testing in Automated Driving Systems Certification. A Longitudinal Dynamics Validation Example
Riccardo Donà, Sándor Vass, Konstantinos Mattas, Maria Cristina Galassi, Biagio Ciuffo
Abstract
The safety validation of Automated Driving Systems (ADSs) needs a combination of tools to ensure testing in a broad range of traffic scenarios. Among the others, virtual testing is expected to play a major role in the future. Differently from other methods, virtual testing allows examining an ADS in complex driving scenarios involving several road users and characterized by any level of criticality in a safe, efficient and effective way. However, before virtual testing can be used in the ADS certification process, proper validation methodologies have to be established to ensure the appropriateness of the simulation-generated evidence to the end of establishing a “ <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">virtual proving ground</i> ” regardless of the specific ADS. In this context, the present paper summarizes the results of the virtual environment validation exercise which involved a Vehicle-Hardware-in-the-Loop (VeHIL) setup against real-world experiments. The analysis only embraces the longitudinal dynamics due to the limitation of the virtual environment. Nonetheless, the methodology presented can be straightforwardly extended to different toolchains to cover more advanced ADS for the sake of virtual certification. The manuscript has a twofold contribution. On one side, it gives a quantitative estimation of the fidelity level achievable using a state-of-the-art VeHIL environment, thus standing apart from validation activities purely based on qualitative comparisons and contributions mostly concerned with the validation of the ADS itself. Secondly, it provides an end-to-end validation procedure that could be generalized to other cases of study or different testing setups. The results achieved are encouraging as they show an overall good match between real-world and simulated data. However, open issues remain in order to define a complete validation framework for virtual testing environments.