Abstract: Taking the evolutionary background of driverless technology as the starting point, this paper reviews the concept of autonomous navigation maps, i.e., high definition maps or HD maps, along with their content components, data sources and differences from ordinary navigation maps. On this basis the author introduces the main modules of the HD mapping process, including data acquisition and update of the a priori map layer, real-time construction of the dynamic map layer, edge computing, data standardization and compression services. Four main research bottlenecks are highlighted: ① automatic modeling of complex driving environments; ② data acquisition for corner cases; ③ navigation decisions in ethical dilemmas; ④ safety performance evaluation of driverless technologies. These bottlenecks are challenging the limits of autonomous navigation capability of HD maps from different perspectives, and demand a global collaboration among government legislatures, related enterprises, research institutions, and the public. With the continuous improvement of driverless technology, the focus of safety-first research will gradually expand into a mobile experience that balances safety and trustworthiness, environmental protection and energy saving, and physical and mental pleasure. The structured digital code of HD maps will be customized to a diverse range of high legibility maps for the human eye. How to leverage the cost-effectiveness of navigation data, generate high precision maps and high legibility maps in real time, and provide optimal services for both machines and humans will become a new research hotspot.

Key words: driverless vehicles, high definition maps, edge computing, behavior prediction, corner cases, ethical dilemmas