Project

Full title: Cost and Resource Efficient Data Collection (CoREDaC)

The CoREDaC project aims to develop a cost- and resource-efficient data acquisition technology for Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD) development. Current fleet-based data collection practices generate massive volumes of image data, leading to high costs for transfer, storage, annotation, and processing, while still leaving important safety-critical scenarios insufficiently covered. Building on results from the previous DeVeLop project, which identified redundancy and similarity-induced data leakage in automotive perception datasets, CoREDaC shifts the focus from post-hoc dataset cleaning to acquisition-time control. The project will establish measurable scenario coverage indicators, develop runtime methods for similarity and novelty estimation, and integrate these into a closed-loop acquisition control framework. This framework will guide whether image data should be logged, retained, or filtered based on both coverage needs and redundancy levels. Through computational experiments and industrially relevant validation with Volvo Cars, Zenseact, and the University of Gothenburg, the project will demonstrate how guided acquisition can reduce low-informative data while preserving safety-relevant scenario coverage and validation credibility. The expected outcome is a practical, scalable, and safety-informed data acquisition layer that supports more efficient, sustainable, and trustworthy AD/ADAS development.

Traffic safety benefit: The project improves traffic safety by enabling AD/ADAS systems to be validated with more representative coverage of safety-critical scenarios across the intended ODD, reducing the risk of overconfident performance estimates caused by redundant or biased data. By guiding data acquisition toward informative and underrepresented traffic situations, it strengthens the evidence base for safer automated driving functions.

Key words: AD/ADAS, autonomous driving, cost-efficient data collection, scenario coverage, Operational Design Domain, runtime data acquisition, redundancy reduction, similarity estimation, data leakage, safety assurance, automotive perception, closed-loop acquisition control.