Skip to content

Latest commit

 

History

History
60 lines (49 loc) · 3.04 KB

README.md

File metadata and controls

60 lines (49 loc) · 3.04 KB
Raw

Reason2Drive: Towards Interpretable and Chain-based Reasoning for Autonomous Driving

[Paper]

Reason2Drive: Towards Interpretable and Chain-based Reasoning for Autonomous Driving,
Ming Nie, Renyuan Peng, Chunwei Wang, Xinyue Cai, Jianhua Han, Hang Xu, Li Zhang
Arxiv preprint

Introduction

This is the official implementation of Reason2Drive: Towards Interpretable and Chain-based Reasoning for Autonomous Driving. We introduce Reason2Drive, an autonomous driving benchmark with over 600K video-text pairs, aimed at facilitating the study of interpretable reasoning in complex driving environments.

Annotation Schema

img|center We first leverage a diverse array of publicly available datasets, including nuScenes, Waymo, and ONCE, and then parse their comprehensive object metadatas into JSON-structured entries. Each object entry contains various details pertaining to its driving actions. Afterwards, these extracted entries are filled into predefined templates, which are divided into different tasks (i.e., perception, prediction and reasoning) at both object-level and scenario-level. Subsequently, GPT-4 and manual annotations are involved for verification and enrichment purposes.

Dataset Stats

img|center Reason2Drive dataset stands as the largest dataset to date, surpassing others in terms of both dataset size and the inclusion of extensive long-text chain-based reasoning references. img|center We split the dataset according to the task (perception, prediction and reasoning) and target. The benchmark exhibits a balanced distribution.

Methodology

img|center The pipeline of our proposed framework. The highlighted yellow box and red curve in the perception result image respectively represent the visualization of predicted location and motion.

Data Preparation

Download links

We provide download links of reason2drive_mini in both google drive and baidu disk for facility:

Google drive: link

Baidu disk: link

TODO

  • Release benchmark (mini-version)
  • Release benchmark (full-version)
  • Release evaluation code
  • Release training code

BibTeX

If you find our work useful in your research, please consider citing our paper:

@article{nie2023reason2drive,
  title={Reason2Drive: Towards Interpretable and Chain-based Reasoning for Autonomous Driving},
  author={Nie, Ming and Peng, Renyuan and Wang, Chunwei and Cai, Xinyue and Han, Jianhua and Xu, Hang and Zhang, Li},
  booktitle={arXiv preprint},
  year={2023}
}

Acknowledgements

We thanks for the opensource projects.