Automated ADAS Testing using Automata Learning - A Case Study in Learning-Based Testing

This repository contains the evaluation code for the paper "Automated ADAS Testing using Automata Learning - A Case Study in Learning-Based Testing". We evaluated three testing methods for an adaptive cruise control.

Setup

The provided setup script setup.sh can be used to install the dependencies necessary for running the evaluation code. It assumes an existing python 3.10 installation. In short it

• downloads and builds version 2.31.7 of the esmini simulator
• creates links for the resulting shared libraries and other resources
• creates a virtual environment in the folder venv and installs the python requirements.

Running

The file main.py provides methods for running experiments and provides an example for their use. In particular, the method run_method(method, seed) runs the specified method using default parameters and the given random seed. It returns a triple sul, model, stats, where sul is the interface to the system under learning, model is the behavioral model learned during testing and stats is a dictionary containing various KPIs. The test cases that were used in the run are contained in sul.io_sequences.

Acknowledgements

This work uses the esmini simulator for OpenSCENARIO XML and the AALpy automata learning library. The python bindings for esmini are generated based on ctypesgen.