Function-Driven Generation Method for Continuous Scenarios of Autonomous Vehicles

The scenario-based test method is now drawing more and more attention in the field of the test for autonomous vehicles. The predefined scenarios are used in the safety verification and performance evaluation of autonomous vehicles. However, the traditional generation method for predefined scenarios is parameterized and open-looped, which makes it challenging to generate diverse and complex scenarios. It is critical when testing high-level autonomous vehicles to verify their reliability in multiple behavior transitions. In this paper, a generation method for the continuous scenario is proposed to realize a function-driven iteration of scenarios for autonomous driving systems (ADS). The method consists of a functional model of ADS and a formal description of abstract scenario. Among them, the functional model is introduced to describe the autonomous driving functions and serve as a simplified decision-making process based on rules to decide which actions should be taken under certain scenarios. The formal description of abstract scenario based on the grid and matrix is developed to describe scenarios formally and consistently at a semantic level, supporting the mathematical computation of scenario transition and risk degree. Finally, by selecting different actions of surrounding traffic participants and iterating them with the functional model, continuous scenarios can be generated from a specific initial scene. The method is applied to the intelligent driver model (IDM) and 1000 continuous scenarios are generated randomly, 76 of which are filtered according to the ego vehicle’s behaviors and analyzed. The results demonstrate that continuous scenarios can trigger the ego vehicle’s behavior sequences under complex situations to verify its reliability. Compared with existing methods for generating predefined scenarios, ours improve the diversity and fulfill multiple interactions between the ego vehicle and surrounding vehicles.