A Comparative Study of Risk Field Modeling for Interactive Driving in Mixed Traffic

(repo under continuous updating)

This interim repo contains packages for modelling of state-of-the-art Interaction Fields with Social Value Orientation based on various BEV datasets (My sample exiD dataset has been archived in "C:\exiD-tools\data", so the data directory needs to be tailored for other users)

GVF Modeling

The Gaussian Velocity Field (GVF) is a mathematical model used to represent interactions between multiple vehicles during lane-change scenarios. The GVF is defined in a Region of Interest (ROI) around an ego vehicle and captures the relative velocity field generated by surrounding vehicles. referenced: "Spatiotemporal learning of multivehicle interaction patterns in lane-change scenarios" (IEEE-TITS)

Key Features:

• Data Loading: Parse exiD dataset CSV files (tracks, tracksMeta, recordingMeta)
• GVF Construction: Build velocity fields using Gaussian Process regression
• Visualization: Plot GVF with vehicle positions and background images
• Logging: Save GVF data in multiple formats (NPZ, CSV, JSON)
• Analysis: Compute features and statistics from velocity fields

visualization of Gaussian Velocity Fields of ego truck and surrounding vehicles:

python exid_gvf_svo_visualization.py --data_dir "C:\exiD-tools\data" --recording 25
#sample usage with logged csv recording the occlusion scenario in exiD:
python exid_gvf_svo_visualization.py --data_dir C:\field_modeling\data\exiD --occlusion-csv "C:\field_modeling\src\output_roles\rec35_ego13_frame140\occlusion_log.csv" --occlusion-row 5 --occlusion-ego-role blocked --output_dir ./output_occlusion_gvf_exid

Mechanical Wave (Aggressiveness) Modeling

visualization of Aggressiveness Distributions of ego truck and surrounding vehicles:

python exid_mech_svo_visualization.py --data_dir "C:\exiD-tools\data" --recording 25

Group Aggressiveness Modeling

visualization of Aggressiveness Distributions of ego truck and surrounding vehicles (aggregated as a group):

python exid_mech_group_visualization.py --data_dir "C:\exiD-tools\data" --recording 25

APF Modeling

visualization of mutual SVO with symmetric evaluations:

python exid_enhanced_svo_visualization.py --data_dir "C:\exiD-tools\data" --recording 25 --output_dir "./enhanced_output"

visualization of SVO with APF in selected recording frame (example: 25):

python exid_svo_apf_visualization.py --data_dir "C:\exiD-tools\data" --recording 25 --output_dir "./output"

EDRF Modeling

integrating deep learning-based multimodal trajectory prediction results with Gaussian distribution models to quantitatively capture the uncertainty of traffic entities’ behavior (ITSC 2024)

python exid_erdf_visualization.py --data_dir C:\field_modeling\data\exiD --occlusion-csv "C:\field_modeling\src\output_roles\rec35_ego13_frame140\occlusion_log.csv" --occlusion-row 5 --occlusion-ego-role blocked --output_dir ./output_occlusion_gvf_exid --light-theme

car interaction visualization (enhanced version, pending frame enhancements):

# Interactive visualization (main program)
python exid_optimized_visualization.py --data_dir /path/to/exid/data --recording 25

# Static analysis plots
python exid_corrected_svo_visualization.py --data_dir /path/to/exid/data --recording 25

Bayseian Survival Analysis:

referenced: "Bayesian survival analysis of interactions between truck platoons and surrounding vehicles through a two-dimensional surrogate safety measure"

python exid_bayesian_platoon.py --data_dir C:\exiD-tools\data --recording 25 --output_dir ./your_defined_subfolder

Classification of agent roles and occlusion analysis

python exid_role_occlusion_analysis.py --data_dir C:\exiD-tools\data --recording 25  --ego_id 101 --output_dir ./your_defined_subfolder
# sample usage:
python exid_role_occlusion_analysis.py --data_dir C:\field_modeling\data\exiD --recording 32 --no-animation --display-neighbors 3 --max-ego-occlusions 3 --align-horizontal

The following programs need the activation of "drone-dataset-tool38":

cd "C:\exiD-tools\data"
cd src
conda activate drone-dataset-tools38

detailed instructions found at: https://github.com/zxc-tju/exiD-tools/tree/master

Dataset structure:

C:\exiD-tools\data\
├── 00_tracks.csv
├── 00_tracksMeta.csv
├── 00_recordingMeta.csv
├── 00_background.png
├── 01_tracks.csv
├── ...
├── 25_tracks.csv          ← recording
├── 25_tracksMeta.csv      ← Vehicle metadata
├── 25_recordingMeta.csv   ← Recording info
├── 25_background.png      ← Aerial image
├── ...
└── Maps/
    ├── location1.osm      ← Lanelet2 HD map
    └── location1.xodr     ← OpenDrive HD map

---

Citation of the exiD dataset:

@inproceedings{exiDdataset,
               title={The exiD Dataset: A Real-World Trajectory Dataset of Highly Interactive Highway Scenarios in Germany},
               author={Moers, Tobias and Vater, Lennart and Krajewski, Robert and Bock, Julian and Zlocki, Adrian and Eckstein, Lutz},
               booktitle={2022 IEEE Intelligent Vehicles Symposium (IV)},
               pages={958-964},
               year={2022},
               doi={10.1109/IV51971.2022.9827305}}