Chat2Scenario: Scenario Extraction From Dataset Through Utilization of Large Language Model

🚨 Important News 🚨

June 17, 2024: The web app "chat2scenario" is accessiable via https://chat2scenario.streamlit.app/

July 20, 2024: ASAM OpenSCENARIO 1.0, 1.1, and 1.2 are available. IPG CarMaker text format is deprecated, as CarMaker has been compatible with ASAM OpenSCENARIO since version 13.0,

Description

Chat2Scenario is a web-based tool that allows users to search for specific driving scenarios within a dataset by inputting descriptive functional scenario text. Results are presented in ASAM OpenSCENARIO and IPG CarMaker .txt format. For more details of Chat2Scenario implementation, please refer to our paper here.

Citation

@INPROCEEDINGS{10588843,
  author={Zhao, Yongqi and Xiao, Wenbo and Mihalj, Tomislav and Hu, Jia and Eichberger, Arno},
  booktitle={2024 IEEE Intelligent Vehicles Symposium (IV)}, 
  title={Chat2Scenario: Scenario Extraction From Dataset Through Utilization of Large Language Model}, 
  year={2024},
  volume={},
  number={},
  pages={559-566},
  keywords={Measurement;Intelligent vehicles;Large language models;Natural language processing;Testing;Large Language Model;Scenario Extraction;Automated Driving Systems;Virtual Testing},
  doi={10.1109/IV55156.2024.10588843}}

Usage

Without installation

You can directly access the web app via the following link: https://chat2scenario.streamlit.app/

With installation

Follow the installation and operation instructions to download the source code and use it locally.

Installation

Pre-requirements

Python installation
- This repository is developed under python 3.9.19 in Anaconda.
Application for dataset
- The highD dataset can be applied via (https://levelxdata.com/highd-dataset/).
OpenAI API key
Clone this repository
```
git clone https://github.com/ftgTUGraz/Chat2Scenario.git 
```
Install all required packages

Method 1 - in your original environment
```
pip install -r requirements.txt
```
Method 2 - create a new environment through yml file
```
conda env create -f environment.yml
conda activate chat2scenario
```

Operation

Open the Project
- Open the entire project using a code editor e.g., Visual Studio Code
- Run the project in the terminal with the following command
```
streamlit run .\Chat2Scenario-Web.py
```
- The Chat2Scenario web application will appear
Set Up Everything
- Select "highD" dataset option (currently, only highD is compitable).
- Upload the corresponding dataset (only the xx_tracks.csv file should be uploaded).
- Select a criticality metric and specify the threshold.
- Enter your OpenAI API Key.
- Choose the desired scenario format.
- Provide a descriptive text for the scenario e.g., "The ego vehicle maintains its lane and velocity. Initially, Target Vehicle #1 is driving in the left adjacent lane. It then accelerates and changes lanes to the right, eventually driving in front of the ego vehicle."
Run
- Click "Preview searched scenario" to search for and preview scenarios.
- Scenarios are ready for download when the prograss bar reaches 100%.
Download
- Click "Extract original scenario" to start extraction process.
- After the progress bar reaches to 100\%, click "download" to acquire the xosc files.
(Optional) Modify the path
- If necessary, modify the path of RoadNetWork in the generated OpenSCENARIO.
```
<RoadNetwork>
<LogicFile filepath="../xodr/highD_01_highway.xodr"/>
</RoadNetwork>
```
Visualization
- Replay the xosc in Esmini based on the userguide
- Replay the txt file in CarMaker based on the Q&A

Contributing

Contributions are what make the open source community such an amazing place to be learn, inspire, and create. Any contributions you make are greatly appreciated.

Fork the project.
Create your feature branch (git checkout -b feature/enhancement)
Commit your changes (git commit -m 'add some enhacement')
Push to the branch (git push origin feature/enhancement)
Open a pull request

Developers

Yongqi Zhao (yongqi.zhao@tugraz.at)
Wenbo Xiao (wenbo.xiao@student.tugraz.at)

For help or issues using the code, please create an issue in this repository or contact Yongqi Zhao at yongqi.zhao@tugraz.at, contact Wenbo Xiao at wenbo.xiao@student.tugraz.at. You can also ask questions in our SLACK Workspace. Click here(renewed link)!

Acknowledgement

This work was supported by FFG in the research project PECOP (FFG Projektnummer 893988) in the program "Bilateral Cooperation Austria - People's Republic of China/Most 2nd Call".
The codebase is built upon other these previous codebases:

ftgTUGraz / Chat2Scenario

readme