An article from Dongguk University in South Korea, 2020, is relatively new. It mainly introduces a new CNN-based agent selection method and a simulation scenario generation method. It also introduces an event-centric distribution mechanism.
Personally, I feel that the quality of the article is average, and it does not describe many key points, and it is a bit confusing.
When generating scene factors, it is good to consider generating near the agent, which is useful for reference and can make the traffic flow more complex and realistic.
The experiment and comparison link in the article is mainly for comparison with SVM.