A Real-Time, Tiled C++ Runtime Library for High-Definition Autonomous Vehicle Maps: Implementation and Evaluation

Date

2026-9-4

Author

Özen, Salih Namık

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

47
views

0
downloads

Cite This

Autonomous driving requires high-definition (HD) map information that can be accessed efficiently for localization, routing, path construction, and trajectory generation. Although detailed road networks can be constructed offline, their in-vehicle use imposes additional requirements on memory consumption, data access, object lifetime, and execution time. This thesis presents \emph{RoadNet-HD}, a C++23 runtime library for the online use of a previously constructed analytical RoadNet representation based on clothoid curves. RoadNet-HD preserves the hierarchical Segment--Node--Junction road topology and LaneGroup--Lane geometry with explicit LaneConnections. Persistent RoadNet data are stored in compact binary chunks, while a vehicle-centered working set controls runtime materialization. Bounded caching, cross-chunk resolution, spatial indexing, and staged LaneGroup materialization keep road topology accessible while instantiating analytical lane geometry only where required. The runtime supports spatial queries, pose-to-lane localization, Dijkstra-based road routing, lane-level route resolution, continuous path construction, clothoid-based lane changes, speed-profile generation, and time-parameterized trajectories. Evaluation on five highway, interchange, and dense urban RoadNet configurations confirms exact preservation of persistent topology and numerical agreement between MATLAB and C++ analytical geometry. Controlled localization retains the source lane for 99.9996% of 234,468 test poses, and representative experiments verify routing and path construction across junctions and chunk boundaries. Persistent loading is substantially more expensive than resident-data operations, supporting its separation from the repeated query path. Staged materialization reduces peak memory consumption by factors of approximately 6.8--11.3, demonstrating the scalability benefit of separating persistent map storage from active computational geometry.

Subject Keywords

High-Definition Maps, Autonomous Driving, Runtime Map Representation, Spatial Data Management, Spatial Indexing

URI

https://hdl.handle.net/11511/121014

Collections

Graduate School of Natural and Applied Sciences, Thesis