Python Scripts for the INTERACTION dataset

• these scripts assist you in processing and visualizing the Interaction Dataset
• for details about and access to the dataset, visit https://interaction-dataset.com/

Required Python Packages

• see requirements.txt
• install them with $ pip install -r requirements.txt

Usage

• copy/download the INTERACTION dataset into the right place
  • copy/download the track files into the folder recorded_tracks, keep one folder per scenario, as in your download
  • copy/download the maps into the folder maps
  • your folder structure should look like in folder-structure.md
• to visualize the data
  • run ./main_visualize_data.py <scenario_name> <trackfile_number (default=0)> from folder python to visualize scenarios
• if you only want to load and work with the track files
  • run ./main_load_track_file.py <tracks_filename> from folder python to load tracks

Test Usage without Dataset

• to test the visualization
  • run ./main_visualize_data.py .TestScenarioForScripts
• to test loading the data
  • run ./main_load_track_file.py ../recorded_trackfiles/.TestScenarioForScripts/vehicle_tracks_000.csv
  • or run the unittests python -m unittest

Participating the INTERPRET challenge

We have organized a trajectory prediction challenge (the INTERPRET challenge) based on the INTERACTION data.

The INTERPRET challenge includes three tracks:

• Regular track: prediction models are trained based on the released data. The test set in this track was sampled from the same traffic scenarios (e.g., maps and traffic conditions) as the released data. Observations of the test set are provided, and the prediction results will be submitted as csv files for performance evaluation.
• Generalizability track: prediction models are trained based on the released data. The test set in this track was sampled from different and new traffic scenarios (e.g., maps and traffic conditions) compared to those in the released data. Observations and maps of the test set are provided, and the prediction results will be submitted as csv files for performance evaluation.
• Closed-loop track: this track aims to evaluate the performance of prediction algorithms in a "prediction-->planning" pipeline so that the impacts of predictors to the closed-loop performance can be evaluated. Prediction models are trained based on the released data, and the models will be submitted as docker images. The submitted predictors, whose structures and parameters cannot be accessed by us, will be run in our simulator for "ego vehicles" in many challenging and interactive scenarios selected from the same traffic scenarios as the release data to evaluate their performance. Virtual, dynamic, and reactive agents are included in the simulator as the surrounding entities of the ego vehicle for each scenario.

For detailed instructions for the "regular" and "generalizability" tracks, please visit INTERPRET_challenge_regular_generalizability_track.

For detailed instructions for the "Closed-loop" track, please visit INTERPRET_challenge_Closed_loop.