Skip to content
This repository has been archived by the owner on Feb 3, 2023. It is now read-only.

The code of our paper "Misbehaviour Prediction for Autonomous Driving Systems", including our improved Udacity simulator

License

Notifications You must be signed in to change notification settings

testingautomated-usi/selforacle

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Misbehaviour Prediction for Autonomous Driving Systems

This repository contains the code attached to the paper "Misbehaviour Prediction for Autonomous Driving Systems" by A. Stocco, M. Weiss, M. Calzana, P. Tonella, to be published in the proceedings of the 42nd International Conference in Software Engineering (ICSE 2020).

A preprint of the paper can be found on arXiv.

Dependencies

Software setup: We adopted the PyCharm Professional 2019.3.1, a Python IDE by JetBrains.

If you have anaconda or miniconda installed on your machine, you can create and install all dependencies on a dedicated virtual environment, by running the following command:

# Use TensorFlow with GPU
conda env create -f code-sdc/conda-env.yml

Alternatively you can manually install the required libraries (see the contents of the conda-env.yml files) using pip.

Hardware setup: Training the DNN models (self-driving cars and anomaly detectors) on our datasets is computationally expensive. Therefore we recommend using a machine with GPU. In our setting, we ran our experiments on a machine equipped with a i9 processor, 32 GB of memory, and an Nvidia GPU GeForce RTX 2080 Ti with 11GB of dedicated memory.

Repository Structure

The repository is structured as follows:

- code-predictors
  | This folder contains all the code to run the training and evaluation of the misbehavior predictors.
- code-sdc
  | This folder contains all the code to train the self-driving car models and record them when they are driving.

Other Artefacts

We made the following artifacts available as a torrent here. The files have a combined size of 25 GB (compressed: 7.93 GB) , and the torrent contains:

- Trained SDC Models
  | Contains the weights of the trained self driving car models.
- Trained Failure Predictor Models
  | Contains the weights of the failure predictor models (i.e., the autoencoders and the deeproad pca).
- Training Dataset & Evaluation Dataset
  | These folders contain the datasets of images used for training and evaluation.
- Evaluation Results
  | This folders contains the sqlite databases containing the results of the evaluation.

Usage

Reproduce the results

python code-predictors/evaluation_runner.py

Note: You do not need the simulator to reproduce the results.

Run the pretrained SDC models

Start up our Udacity self-driving simulator, choose a scene, and press the Autonomous Mode button. Then, run the model as follows:

python code-sdc/drive.py <sdc-model>.h5

Train a SDC model

You'll need the data folder which contains the training images.

python code-sdc/train_self_driving_car.py

This will generate a file model-<epoch>.h5 whenever the performance in the epoch is better than the previous best. For example, the first epoch will generate a file called model-000.h5.

Train an anomaly detector

You'll need the data folder which contains the training images.

python code-predictors/training_runner.py

Improved Udacity simulator

Our improved Udacity simulator is available as binary file for Windows and macOS platforms here.

Screenshots and video of the simulator

Note: Click on the screenshot images to watch a demo video.

Reference

If you use our work in your research, or it helps it, or if you simply like it, please cite SelfOracle in your publications. Here is an example BibTeX entry:

@inproceedings{2020-icse-misbehaviour-prediction,
	title= {Misbehaviour Prediction for Autonomous Driving Systems},
	author= {Andrea Stocco and Michael Weiss and Marco Calzana and Paolo Tonella},
	booktitle= {Proceedings of 42nd International Conference on Software Engineering},
	series= {ICSE '20},
	publisher= {ACM},
	pages= {12 pages},
	year= {2020}
}

License

MIT. See the LICENSE.md file.

Contacts

For any questions, feel free to contact Andrea Stocco ([email protected]) or Michael Weiss ([email protected]).

About

The code of our paper "Misbehaviour Prediction for Autonomous Driving Systems", including our improved Udacity simulator

Resources

License

Stars

Watchers

Forks

Languages