PRIBOOT

PRIBOOT is a new data-driven expert system designed to tackle CARLA Leaderboard 2.0. This repository contains the implementation details and instructions for setting up and running PRIBOOT, as described in the paper:

PRIBOOT: A New Data-Driven Expert for Improved Driving Simulations

Daniel Coelho, Miguel Oliveira, Vítor Santos, and Antonio M. López

If you find our work useful, please consider citing:

@article{coelho2024priboot,
  title={PRIBOOT: A New Data-Driven Expert for Improved Driving Simulations},
  author={Coelho, Daniel and Oliveira, Miguel and Santos, Vitor and Lopez, Antonio M},
  journal={arXiv preprint arXiv:2406.08421},
  year={2024}
}

Setup

1. Clone the repository with git clone [email protected]:DanielCoelho112/priboot.git

2. Download the folder birdview_cache and place it in the root directory of the PRIBOOT repository.

3. Create a folder to store the results. In that folder place the folder priboot_original which contains the weights of the model.

4. Download CARLA 0.9.15.

5. Run the docker container with docker run -it --gpus all --network=host -v results_path:/root/results/priboot -v priboot_path:/root/priboot danielc11/priboot:0.0 bash where results_path is the path where the results will be written, and priboot_path is the path of the PRIBOOT repository.

Evaluating the agent

1. Start the CARLA server

2. Open the file priboot/config/priboot_original/experiment.yaml and configure the various options according to your requirements.

3. Run: python3 priboot/leaderboard/leaderboard/leaderboard_evaluator.py -en priboot_original