Copula-Based Interactive Behavior Modeling for Naturalistic Test Scenarios Generation.

With the rapid development of autonomous driving technology, ensuring the safety of Autonomous Vehicles (AVs) has become a paramount concern. To tackle these safety challenges and enhance testing efficiency, this study presents an innovative method for generating naturalistic simulation scenarios for AV testing. The proposed method perceives the actions of the driving model as a probability problem, thus enabling the construction of a probability description model for naturalistic interactive behavior. Utilizing Naturalistic Driving Data (NDD) from DiDi autonomous driving dataset, the Copula function is employed to establish a high-dimensional, multivariate joint distribution model of naturalistic behavior variables. This constructed model is designed to encapsulate the distribution characteristics of interactive behavior variables effectively. This approach allows for the generation of behaviors featuring naturalistic interactions and driving scenarios that are unbiased reflections of real-world conditions. Experimental validation corroborates the effectiveness of these constructed simulation scenarios in emulating real-world driving behaviors. A comparative analysis with simulation scenarios based on the Intelligent Driver Model indicates a substantial reduction in the Mean Squared Error of vehicle variable distribution relative to real data in the Copula-based simulation scenarios, thereby validating the superiority of our proposed model.