Datasets and data spaces in the autonomous driving domain: insights and trends

Abstract

The rapid evolution of autonomous driving technologies is deeply reliant on high-quality datasets and robust data management solutions that cannot be achieved by a single actor working in isolation. This paper examines the growing role of datasets and data spaces in the mobility domain, providing a conceptual analysis of current approaches and identifying key challenges and emerging trends that inform the design of an autonomous vehicle data space framework. While peer-to-peer solutions for data sharing already exist, the concept of data spaces can significantly scale these approaches, creating a self-improving feedback loop that enhances the performance of AI-based driving models by matching the demand for data with the data offered by providers. Specifically, the paper addresses the challenge of acquiring intra-vehicular data to enhance assisted driving functionalities and proposes a methodology for building a data space specialized for automated driving applications. Finally, the paper explores representative use cases, including a Level 4 automated shuttle, TalTech iseAuto, Estonia, and an all-terrain vehicle provided by CERTH, Greece, by conceptualizing data sharing across different domains, where data providers and consumers exchange information within a data space framework, driving continuous improvement in model performance.