Source code for the following paper:

Chen, Xianda, et al. "FollowNet: A Comprehensive Benchmark for Car-Following Behavior Modeling."

This notebook demonstrates how to achieve the car following models from traditonal models to data driven models. Motivation: given extracted car following events from five open datasets with the same data formate and train the car follow models. Author: Chen Xianda.

The extracted car following events are avaliable for download. Provide a tutorial of the data format and how to run the traditional models and the data-driven models.

Extracted car-following events are stored in data/ folder. The colab tutorial takes the highD data for experiments first.

The datasets are HighD, SPMD(DAS1, DAS2), Waymo, Lyft, NGSIM. Each has its own training, validation and test part.

Run the colab notebook directly! Details are in the notebook below.

Pretrained models are stored in trained_model/ folder.

Below is the average time gap during car following (s). For more results stored in results/ folder.

Collsion rate

MSE of spacing

[email protected]

[email protected]

If you use extracted car following data / FollowNet in your own work, please cite:

@misc{chen2023follownet,
      title={FollowNet: A Comprehensive Benchmark for Car-Following Behavior Modeling}, 
      author={Xianda Chen and Meixin Zhu and Kehua Chen and Pengqin Wang and Hongliang Lu and Hui Zhong and Xu Han and Yinhai Wang},
      year={2023},
      eprint={2306.05381},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}