Towards a new multi-representation LOD model for high-definition maps in automated driving

Recent advances in automated driving (AD) and intelligent connected vehicles yield two paradigms for environmental understanding, namely map-based approaches and sensor-based perception. Instead of being mutually exclusive, the two paradigms are complementary and supply redundant spatial knowledge to improve safety and situational awareness. Existing map models are either limited in accommodating different levels of map details, including standard-definition (SD) maps, advanced driver-assistance system (ADAS) maps, and high-definition (HD) maps, or fail to incorporate real-time environmental perception. This paper proposes a novel representation model to enrich map details and bridge the gap between perception and maps: the multi-representation LOD (Level of Detail) model. It enables the integration of perception data and map features within a unified, feature-based representation framework. The LOD concept is extended to accommodate varying map details ranging from SD to HD maps. The resulting high-definition map language (HDML) complies with ISO and OGC standards. It facilitates interoperable data exchange and supports map update learning for intelligent transportation and AD applications.