Code for "Beyond behaviour change: investigating alternative explanations for shorter time headways when human drivers follow automated vehicles"

This study is published in the journal "Transportation Research Part C: Emerging Technologies" with gold open access, available at https://doi.org/10.1016/j.trc.2024.104673.

Abstract

Integrating Automated Vehicles (AVs) into existing traffic systems holds the promise of enhanced road safety, reduced congestion, and more sustainable travel. Effective integration of AVs requires understanding the interactions between AVs and Human-driving Vehicles (HVs), especially during the transition period in which AVs and HVs coexist in a mixed traffic environment. Numerous recent empirical studies find reduced headways of human drivers following an AV compared to following an HV, and attribute this reduction to behaviour changes of following vehicle drivers. However, more factors may be at play due to the inherent inconsistencies between the comparison conditions of HV-following-AV and HV-following-HV. This study scrutinises three alternative explanations for the observed reduction in headways: (1) systematic differences in car-following states during data collection, (2) systematic differences in driving variability between leading AVs and HVs, and (3) systematic differences in driving characteristics of leading AVs versus HVs. We use a large-scale dataset extracted from Lyft AV motion data, and isolate each of these explanations by data stratification and simulation. Our results show that all three mechanisms contribute to the observed reduction in headways of human drivers following AVs. In addition, our findings highlight the importance of driving homogeneity and stability in achieving reliably shorter headways. Thereby, this study offers a more comprehensive understanding on the difference between HV-AV and HV-HV interactions in mixed traffic, and is expected to promote more effective integration of AVs into human traffic.

Package requirements

jupyter notebook, zarr, numpy, pandas, pytables, tqdm, matplotlib, scipy, joblib, pytorch

In order to repeat the experiments:

Step 0. Preparation

Create a conda environment for repeating the experiments. Install the required packages as listed above.

Clone this repository, then either 1) create/define a folder for data saving and copy the subfolders in "Data"; or 2) use the folder "Data" directly.

Step 1. Download and save data

Download the trajectory data of Lyft from https://github.com/RomainLITUD/Car-Following-Dataset-HV-vs-AV and save them in the folder "Data/InputData/Lyft/"; download processed data of Waymo from https://data.mendeley.com/datasets/wfn2c3437n/2 and save it (all_seg_paired_cf_trj_final_with_large_vehicle.csv) in the folder "Data/InputData/Waymo/".

Step 2. Standardise data format (of Waymo and Lyft)

Run data_format_standardization.py in the current parent folder first to preprocess the trajectories.

Step 3. Regime categorisation

In the folder "Car-following regime categorisation", use regime categorisation.ipynb to categorise car-following regimes in Waymo and Lyft datasets.

Step 4. IDM calibration and simulation

In the folder "Car-following modelling and simulation", run idm_data_selection.py to select car-following pairs that cover sufficient regimes for Intelligent Driver Modelling.

Then in the sub-folder "Car-following modelling and simulation/IDM calibration", run idm_calibration.py to calibrate IDMs and run loss_computation.py to calculate calibration loss. Further in the sub-folder "IDM calibration/Appendix", we offer the calibration of the other two car-following models, Newell and Gipps, to enhance the robustness of the results.

Finally, in the sub-folder "Car-following modelling and simulation/Controlled simulation", run cross_follow_leader.py and cross_follow_follower.py to simulate the designed experiments.

Step 5. Leading vehicle classification

In the folder "Leading vehicle driving classification", use dataset_separation.py to separate Lyft data into train, val, and test sets, and then use classifier_lstm.ipynb to train the LSTM classifier, validate the trained model, and save test results.

* In doing regime categorisation, we resued the code from https://github.com/slaypni/fastdtw to apply fast dynamic time warping. This is also indicated in the folder.

* We have run the IDM calibration in Linux with 15 CPUs. To be run on other OSs may need adjustments regarding the number of cores/workers for parallel processing.

Citation

@article{Jiao2024,
  title = {Beyond behavioural change: Investigating alternative explanations for shorter time headways when human drivers follow automated vehicles},
  volume = {164},
  doi = {10.1016/j.trc.2024.104673},
  journal = {Transportation Research Part C: Emerging Technologies},
  author = {Jiao,  Yiru and Li,  Guopeng and Calvert,  Simeon C. and van Cranenburgh,  Sander and van Lint,  Hans},
  year = {2024},
  pages = {104673}
}