Unravelling uncertainty in trajectory prediction using a non-parametric approach

• Visit the DiTTlab demo page below to visualize how the model works:

Dittlab Online Demo (click)

• This is the official source code of the paper manuscript: Unravelling uncertainty in trajectory prediction using a non-parametric approach

• The paper is available on Transportation Research Part C: Emerging Technologies : website link (click)

Quick start

Requirements:

• Python = 3.9
• PyTorch ≥ 1.11
• Shapely = 1.8.5

Packages

• Data preparation needs ROS and lanelet2 toolkit, which only supports Linux systems. Please go to github page to see the installation tutorial

• Before training or testing the model, please install the required packages by:

 pip install -r packages.txt

Data Preparation

• The used INTERACTION dataset can be found by Interaction Webpage.
• The corresponding INTERPRET challenge: Leaderboard.
• Put the INTERACTION data in the correspondings train/val/test folders in interaction_data folder
• Use data_generator.ipynb and mask_generator.ipynb to process the raw data. These two notebooks provide detailed instructions.
• Processed data will be in interaction_merge folder.
• If there is any difficulty in preparing the dataset, please first get the data permission from INTERACTION team and send us an email: G.Li-5@tudelft.nl. We will share the fully-processed data that is ready to use.

Model Training

• Run the TrainingModels.ipynb to train the deep ensembles of UQnet.
• Detailed instructions are provided in the notebook.

Paper Reproduce

• Run the ResultsReproduce.ipynb to get the quantified uncertainty, predictions, etc.
• Here we use the MR minimization strategy, which is the same as the leaderboard.
• Detailed instructions are provided in the notebook.

Reproducing the Accuracy on the Leaderboard

• Run the ForSubmission.ipynb sequentially.
• The generated submission file is exactly the same as shown on the leaderboard

Visualization

• For visualization, please do not use the VisualizeResults notebook. It is the old test.
• Please go to our online interactive demo for visualization (the link is at the top of the page).