Abstract Manager

Autonomous vehicles have been trending over the past decade, however, there are many obstacles that entities are trying to overcome in order to achieve fully autonomous SAE level 5 driving. One of those challenges is the feasibility of efficiently producing, and regularly updating High Definition (HD) maps of existing road infrastructure on a large-scale and in a consistent and seamless manner. HD maps are highly detailed digital twins of road infrastructure that include 3-dimensional representation of road features that are critical to a vehicle’s ability to navigate its path. This includes features such as lane markings, curbs, and road edges. HD maps are produced by first surveying roads using remote sensing techniques such as mobile Light Detection and Ranging (LiDAR) and cameras. In current practice, the collected data is then manually processed to extract HD map features. As a result, large-scale production of HD maps remains a tedious and highly time-consuming task, which is not sustainable when implemented to a large and dynamic network of roads. This work provides insights into Nektar 3D Consulting’s use of mobile LiDAR and AI technology to overcome the aforementioned challenges and produce HD maps in a highly automated manner for the Dubai Municipality. The paper provides an end-to-end illustration of how LiDAR scans of hundreds of kilometres were converted into a simulation ready HD map of the Emirate of Dubai in an automated manner. LiDAR data collected by the municipality was first broken down into features critical to autonomous vehicle navigation using deep learning tools. Unsupervised machine learning and data clustering techniques were then used to refine the linear features in the data before automatically generating the HD map components in a simulation ready format compatible with open source simulation software. Besides illustrating the full process, this work also presents some of the lessons learned and the knowledge gained from such a unique experience of fully automated large-scale HD map production in one of the world’s busiest metropolitans. In summary, this work demonstrates how combining mobile LiDAR technology, AI, and unsupervised machine learning paves the way for sustainable large-scale production of HD maps.