Perception Sensor Model Fidelity Evaluation for Automated Driving System Scenario-Based Simulation Testing

Scenario-based simulation testing has become a mainstream method for performance verification of automated driving systems. However, there is still no unified consensus on how to validate the effectiveness of simulation testing results. In particular, perception sensor models play a central role in mapping simulation scenarios to automated driving systems, and their fidelity evaluation has been receiving increasing attention. This paper proposes a perception sensor model fidelity evaluation method for scenario-based simulation testing of automated driving systems. First, the overall process of fidelity evaluation for perception sensor models in logical scenarios is presented, including the selection of scenarios, evaluation indicators, and the coupling of indicators. Subsequently, multi-layer fidelity evaluation metrics and computational methods are introduced, covering basic layer, sensorium layer, and object layer indices. Next, the analytic hierarchy process and probability distribution methods are used to determine the weights of different fidelity evaluation metrics. The proposed method is validated through LiDAR sensor models constructed using Ray-Tracing technology with different parameters in a front static logical scenario. The fidelity of the four sensor models in this type of scenario is 91.43%, 88.18%, 87.21%, and 81.35%, respectively. The proposed method can be used to verify the process of scenario-based simulation testing for automated driving systems, providing strong support for the validity of automated vehicles.