Fundamental Architectures for High-Integrity Georeferenced Lidar Positioning

In this paper, we classify different approaches, or architectures, for implementing high-integrity, lidar-based positioning for road vehicles. Our emphasis is on analytical, geometry-based algorithms, which we believe offer the strongest opportunity for transparent and rigorous quantitative analysis of system integrity. We identify three important classes of architecture: engineered-target methods that rely on purpose-built signs that can be uniquely identified, landmark-based methods that leverage specific existing infrastructure (e.g., lamp posts or telephone poles), and scan-matching methods that compare the entire scene to a high-definition map. By outlining algorithms developed to test all three architectures, we reveal their similarities and differences. We build on these algorithmic descriptions to establish how the architectures can be complementary, in the sense that the same vehicle might combine all three architectures to achieve higher availability, integrity, accuracy continuity, or resiliency across a range of operating domains. We also identify a wide range of potential adversity modes that introduce challenges for establishing a strong integrity case and outline promising future work needed to mitigate these adversities.