A novel framework for adaptive stress testing of autonomous vehicles in the highway

---

Project Structure

• config Dir containing config files for training/evaluating.
• highway_env Dir containing source code.
• weights: Dir storing trained weights.
• LICENSE: File describing license terms.
• main.py: Main file.
• README.md: This file!

---

Installation

Please following instruction from HighwayEnv for installation.

Config

config consist of:

• env_name: class name of simulation environment.
• num_episodes: number of episode to run reinforcement learning for AST.
• num_steps: number of each step in each episode.
• save_result_csv: path to a csv file to store the recorded collision. The format of each collision is described as below: | #episode | #step | ego_speed | ego_acceleration | ego_ast_action | lane_index | speed_crashed_veh | acceleration_crashed_veh | crashed_lane_index | crashed_ast_action | crashed_front | crashed_distance |
• model_path: path to model file to be saved after training reinforcement learning model for AST.
• reward_file (Optional): path to save training reward for further analysis.
• save_pic: path to folder for exporting scene from simulation for further analysis.
• ttc_threshold: a threshold of time-to-collision that used to calculate propability of collision as in paper.
• ego_collision_weight: the value of lambda.

Train and testing

To train and test a model, simply run

python main.py --config_path config/<YOUR_CONFIG>.yaml

• config_path: Path to trainer config file, containing details for experiment as described above...

---

Citation

If you used the code in this repository or found the paper interesting, please cite it as

@misc{trinh2024,
      title={A novel framework for adaptive stress testing of autonomous vehicles in multi-lane roads}, 
      author={Linh Trinh and Quang-Hung Luu and Thai M. Nguyen and Hai L. Vu},
      year={2024},
      eprint={2402.11813},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2402.11813}, 
}

---

Acknowledgements

• Especially thanks to HighwayEnv with there excellent work. This code base borrow borrow frameworks to accelerate implementation.
• This project is supported by a grant from the Smart Pavements Australia Research Collaboration Hub.

Contact

If you have any problems about this work, please contact Linh Trinh at [email protected].

Licence

This project is licenced under the Commons Clause and GNU GPL licenses. For commercial use, please contact the authors.

---