Automatic Construction of HD Maps for Simulation-Based Testing of Autonomous Driving Systems

Siqi Wang, Changwen Li, Tiantian Sun, Fuqi Jia, Rongjie Yan, Jun Yan

Abstract: In the development and testing of autonomous driving systems, high-definition (HD) maps play an indispensable role by providing vehicles with precise information about their surroundings, encompassing road markings, topographical features, and intricate junction layouts. Despite their importance, creating HD maps through real-world data acquisition or manual design can be laborious, time-consuming, and hard to ensure correctness and diversity. This paper introduces a model-based framework for the automated construction of HD maps. The framework considers junctions as essential components within the road network topology. It formulates the inherent constraints of junction configurations and their interconnections through a Maximum Satisfiability (MaxSAT) model. Based on the optimized configuration for the junctions, the framework defines and integrates the external and internal road connections linked to these junctions, alongside assigning corresponding traffic signal systems. In the experimental evaluation, the comparison with real-world maps and embedded maps in open-source simulation platforms demonstrates that the generated map maintains a compelling balance between diversity and structural simplicity, thus effectively replicating the complexity of real-life road networks. Moreover, the usability of the generated maps has been validated via simulation-based testing of an open-source autopilot system, uncovering several map-related issues.