SCTrans: Scenario Dataset for Simulation Testing of Autonomous Driving Systems

In this work, we introduce a transformation-based approach, namely SCTrans, which utilizes existing real-world traffic scenario datasets to create simulation scenario files. For more approach details, please see our ICSE-24 paper.

To curate an initial version of our scenario dataset, we randomly selected 1,994 valid traffic scenario recording files as data sources, from representative traffic scenario datasets, including CommonRoad[1], inD[2], and highD[3]. Finally, we got 1,994 simulation scenario files and 426 map description files, which are compatible with advanced ADS testing platforms (e.g., LGSVL[4]+Apollo[5] and Carla[6]+Autoware[7]. For ease of using our scenarios, we provide scenario-runner scripts for LGSVL and Carla, available at our Github Repo[8]. Besides, for ease of scenario selection, we manually label all our simulation scenarios according to ISO 21448 (i.e., guiding principles for ADS SOTIF testing). The labeling results are listed in scenario-tag.xlsx.

======== How to Use ========

Please refer to our Github Website. We provide the refined versions of the LGSVL & Carla scenario runners. The scenario-runner directory contains the complete, modified source code along with comprehensive usage instructions.