Generating lane-changing paths for vehicles with specific risk levels based on CGAN

The emergence of autonomous driving vehicles provides a more convenient and comfortable transportation experience for human society, but it also brings some safety hazards. In recent years, due to the inherent limitations of autonomous driving algorithms, self-driving cars have been involved in multiple traffic accidents during lane-changing scenarios. To address this issue, there is a need to propose a method for generating lane-changing trajectories with specific risk levels to assist autonomous driving algorithms in conducting safety tests, thereby enhancing the safety of autonomous vehicles. This paper utilizes a Temporal Convolutional Network (TCN) to extract risk features from lane-changing scenarios and applies an autoencoder along with Conditional Generative Adversarial Networks (cGANs) to learn from and generate lane-changing data with specific risks. Additionally, an evaluation method has been proposed to verify the authenticity, diversity, and similarity of the generated data.