Scenario Diffusion: Controllable Driving Scenario Generation with Diffusion
Abstract
Automated creation of synthetic traffic scenarios is a key part of scaling the safety validation of autonomous vehicles (AVs). In this paper, we propose Scenario Diffusion, a novel diffusion-based architecture for generating traffic scenarios that enables controllable scenario generation. We combine latent diffusion, object detection and trajectory regression to generate distributions of synthetic agent poses, orientations and trajectories simultaneously. This distribution is conditioned on the map and sets of tokens describing the desired scenario to provide additional control over the generated scenario. We show that our approach has sufficient expressive capacity to model diverse traffic patterns and generalizes to different geographical regions.
Cite
Text
Pronovost et al. "Scenario Diffusion: Controllable Driving Scenario Generation with Diffusion." Neural Information Processing Systems, 2023.Markdown
[Pronovost et al. "Scenario Diffusion: Controllable Driving Scenario Generation with Diffusion." Neural Information Processing Systems, 2023.](https://mlanthology.org/neurips/2023/pronovost2023neurips-scenario/)BibTeX
@inproceedings{pronovost2023neurips-scenario,
title = {{Scenario Diffusion: Controllable Driving Scenario Generation with Diffusion}},
author = {Pronovost, Ethan and Ganesina, Meghana Reddy and Hendy, Noureldin and Wang, Zeyu and Morales, Andres and Wang, Kai and Roy, Nick},
booktitle = {Neural Information Processing Systems},
year = {2023},
url = {https://mlanthology.org/neurips/2023/pronovost2023neurips-scenario/}
}