Generating Causal Explanations of Vehicular Agent Behavioural Interactions with Learnt Reward Profiles

The data, code, parameters, and videos here relate to a causal explanation generation approach utilising learnt reward profiles and the experiments evaluating it. Specifically, the framework generates explanations for agents within the autonomous driving domain.

Code

The code for the framework utilised by the methodology provided by the associated paper as well as any utility scripts.

Data

Data and parameters relating to the experiments carried out upon the framework. Note that the quantitative experiments referred to here rely upon the highD dataset which is free for non-commercial use by application at the following link: https://www.highd-dataset.com/. However, as the dataset is not public, only the output experiment data is available in this repository. In addition to the highD dataset, the inD (https://levelxdata.com/ind-dataset/) and exiD (https://levelxdata.com/exid-dataset/) datasets are also utilised in qualitative experiments. Similar to highD these are not open access, but free access can be requested for non-commercial activies.

Videos

Contains the summary video for the paper as a whole, as well as qualitative experiment videos depicting the scenarios discussed in the paper.

Paper

Corresponding conference paper accepted at IEEE International Conference on Robotics & Automation 2025 (Pre-print): TBD

Please cite the following if you use the contents of this repository in your work:

@inproceedings{howard2025generating,
  author    = {Howard, Rhys Peter Matthew and Hawes, Nick and Kunze, Lars},
  booktitle = {2025 IEEE International Conference on Robotics and Automation (ICRA)}, 
  title     = {Generating Causal Explanations of Vehicular Agent Behavioural Interactions with Learnt Reward Profiles}, 
  year      = {2025},
  volume    = {},
  number    = {},
}