LLM-Driven Autonomous Vehicles Can Inherit Human Driver Biases in Pedestrian Yielding: Results and Implications From A New Benchmark
Irem Yoldas, Jie M. Zhang, Odinaldo Rodrigues, Martim Brandao
💥 This repo contains the VLM (Self Consistency) test and the LLM (All Else Being Equal) test of our benchmark in our paper.
Due to space limitations, you can find the graphs that we were unable to include in the paper article here!
In SC test:
- First we ask whether the pedestrian has an intention to cross and is able to cross safely if the vehicle does not stop as Roud-1.
- Ask questions about demographic information and whether the autonomous vehicle will stop or not as Round-2.
Due to GitHub file size limitations, the SC test images that we used are hosted externally.
You can download them here: Google Drive Link
In LLM Prompts file, you can find two questions:
- For generating written scenarios via images
- A question to be added at the end of the written scenarios for the prompt
In json file, you can find every written scenario that we used!
Due to GitHub file size limitations, the AEBE test scenarios are hosted externally.
You can download them here: Google Drive Link
If you experience any issues accessing the data, feel free to contact us.
If you think this repo or paper is useful, do not forget to cite our paper, please: