Publication

Traffic is the sum of road design, road user behavior, and their interactions. Understanding the dynamics is central to making the transport system efficient, safe, and sustainable. Viscando offers Al-enabled data-driven solutions for traffic analysis, safety diagnostics, intelligent traffic control, naturalistic data collection, and extended perception at the very core of smart cities and autonomous driving. This is enabled by the infrastructure sensor OTUS3D, which is used for collection of accurate traffic movement data.
Software quality assurance (QA) must evolve as systems increasingly rely on AI. Building on initial work on simulation-based testing, we set up this pre-study to pave the way for evolutionary algorithms to generate test scenarios in CARLA that stress OTUS3D to its limits. We developed and published a digital model of an intersection in Lindholmen under an open-source software license. The model is available in two formats to allow importing into MathWorks RoadRunner and the open-source simulator CARLA. We also provide a set of trajectories for road agents.
Our project prepares for future search-based software testing of OTUS3D. Based on the trajectories and the environmental conditions, we propose key parameters to configure traffic scenario execution in CARLA. Furthermore, we share source code for test scenario generation using the NSGA-II algorithm. We present intermediate results as a proof-of-concept in this report. In the next two months, our work will continue and the final results will be published in a MSc thesis at Lund University.