Generation of Autonomous Vehicle Testing Trajectories for Cut-In Scenario Integrating Data and Kinematics

The assessment of autonomous vehicles (AVs) represents a pivotal aspect of technological advancement. However, the evaluation of AVs within a simulated environment presents a number of challenges, including ensuring compliance with physical laws and reflecting the naturalness and diversity observed in real-world driving, which makes simulated testing a challenging issue. In view of this, we introduce a novel approach that integrates data-driven methods with kinematic constraints to generate testing trajectories. Our method conceptualizes trajectory generation as a probabilistic process, using a Gaussian distribution to introduce variations. By incorporating fusion technology, this approach can rely solely on initial scene information, thereby enhancing the practicality and convenience of generating testing trajectories for AV applications. An ablation study of kinematic constraints and data is separately conducted, demonstrating the effectiveness of our model.