WUT researchers working on a scenario database for autonomous vehicles

The goal of the project is to develop an original database of test scenarios for autonomous vehicles (AVs) that reflect road conditions specific to Poland. The database will serve as the foundation for testing vehicle perception systems at automation levels L3–L5 according to the SAE J3016 standard and will be made available to interested entities on a non-profit basis to strengthen the potential of the Polish economy.

The publicly accessible database will enable the design, creation, and evaluation of perception systems for vehicles across various levels of automation. The project’s innovativeness stems from the nature of its reference data, which are distinguished by their diversity and high level of detail. The database will include data from multiple sensors: IMU, GPS, cameras, radars, and LiDARs.

After appropriate processing, these data will undergo annotation. In the process of data annotation and quality assessment, the project will employ artificial intelligence algorithms and advanced digital data processing methods, such as data fusion, automatic object detection and tracking, and active learning algorithms. The selected test routes will feature scenarios typical of Polish road environments, including locations with high accident rates. The database will include road sections of all classes and categories.

"The conducted field tests will take into account various road conditions, such as traffic jams, accidents, or roadworks – including those with limited visibility, for example due to fog. The project's implementation will provide a tool essential for the deployment of AVs, and its results will help shape the national development policy in this area. The DARTS project aligns with numerous development strategies, such as National Smart Specialisations (KIS10), and addresses socio-economic needs, enabling the Polish economy to follow new technological trends and avoid development traps," says Prof. Robert Nowak from the Faculty of Electronics and Information Technology. "My role involves software development – here, it's a pipeline that will convert sensor data from cameras, LiDARs, radars into the appropriate formats, ensure data consistency, then enable automatic object detection, the refinement of such automatic detections by an expert (human), and make the whole system available to users, for instance, for training machine learning models."

The DARTS-PL database is intended for engineers, technicians, and companies working on problems related to the control of automated vehicles. It constitutes a unique collection of recordings of real road situations recorded in Poland – containing specially selected road sections that allow for the replication of a wide spectrum of scenarios incorporating Poland's characteristic infrastructure and traffic organization. Recordings are conducted on selected road sections and intersections in various regions of Poland. The sections were chosen based on two criteria: high accident concentration (based on conducted analyses, nine counties were identified where sections with high accident rates in recent years were determined) and representation of characteristic engineering solutions (sections containing engineering structures and/or marking methods that are used in Poland, including tunnel signage, various classes of railway-road crossings, bicycle routes, etc.).

“The consortium partner is the Motor Transport Institute. The partner provides the vehicle, equipment – meaning sensors and a computer for recording data from the sensors installed on the vehicle – and provides the raw data – selects the routes, drives, records. The most time-consuming will be the correction of automatic annotations by a human expert due to the volume of data – 24 frames per second, 20-second recordings, several sensors in each recording: cameras, LiDARs, radars and others, 800 scenarios," emphasizes Prof. Nowak. "The challenge is delivering high-quality annotations, which is associated with an appropriate measurement setup, synchronization, and automation of the annotation process through detection algorithms based on machine learning and computer vision."

The database has recorded natural road traffic encountered in these locations, with care taken to maintain the diversity of object classes present in the recordings. In addition to the scenarios recorded in selected locations, the DARTS-PL database also offers a special collection of recordings containing objects and situations that may pose a challenge for classical object classification algorithms. The goal of creating the DARTS-PL database is to increase the availability of high-quality data for use in research and development work. It can be used both as an element in developing object detection and classification methods in datasets, and in developing motion planning and control algorithms for automated vehicles. The solution has attracted interest from the Ministry of Infrastructure – for standardizing the certification process – as well as from companies creating perception systems for autonomous vehicles for testing their solutions.

The project will run until June 2027. Its first phase will conclude in November 2025 – the outcome will be the delivery of a measurement setup (vehicle + sensors), an annotation pipeline (this is handled by WUT) and confirmation that everything works correctly on several scenarios. The second phase involves recording and performing annotations, and the final stage is making the database available to users.

The DARTS-PL project has been financed by the National Centre for Research and Development under agreement no. GOSPOSTRATEG-VIII/0001/2022. The entity supervising the project's execution is the Ministry of Infrastructure – Department of Transport Strategy.