Spatial Sampling and Integrity in Lane Grid Maps

Sanchez, Corentin; Xu, Philippe; Armand, Alexandre; Bonnifait, Philippe · 2021 · Crossref

DOI: 10.1109/ivworkshops54471.2021.9669257

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of providing reliable, non-misleading spatial information to the decision-making (DM) modules of autonomous vehicles (AVs) in complex urban environments. The authors argue that traditional object lists lack explicit information about free space, while generic occupancy grids do not align with road geometry or handle localization uncertainty effectively. To solve this, the paper introduces the Lane Grid Map (LGM), a tactical-level representation that combines spatial occupancy with along-track distance metrics. The primary goal is to ensure information integrity—specifically preventing the dangerous misclassification of occupied space as free—by optimizing the spatial sampling step of the map cells. The LGM is constructed by defining Areas of Interest (AOI) derived from an HD map’s topological layer, focusing on primary and secondary lanes relevant to the ego vehicle’s path. The road lanes are sampled along their centerlines into contiguous cells, characterized as Free, Occupied, or Hidden. The characterization process utilizes a free-space polygon approach, where cells are classified based on intersections with perceived object polygons and free-space polygons. The paper evaluates the LGM’s performance using real-world data collected from three experimental vehicles (two Renault Zoe and one Renault Master) in Compiègne, France. The setup included a Velodyne LiDAR for perception and a NovAtel IMU/GNSS system for centimeter-level ground truth localization. To assess integrity under uncertainty, the authors simulated localization errors by adding Gaussian noise to the vehicle’s pose and propagated this uncertainty to the object hulls. The results demonstrate that the LGM’s sampling step is a critical parameter for managing integrity risk. The study defines False Negative Rate (FNR) as the rate of occupied cells misclassified as free, which poses a safety hazard. Without uncertainty propagation, the FNR remains low regardless of the sampling step. However, when localization uncertainty is propagated to object hulls, the FNR increases significantly with finer sampling steps (oversampling). The authors show that by increasing the sampling step (aggregating cells), the system can maintain the FNR below a predefined integrity risk threshold. For instance, with a localization noise standard deviation of 0.5 meters, a sampling step of approximately 3.5 meters is required to guarantee the desired integrity level. This approach allows the AV to trade off spatial resolution for confidence, ensuring that the DM module receives robust information despite sensor and localization errors. The significance of this work lies in its formalization of integrity management for spatial representations in autonomous driving. By linking the spatial sampling step directly to localization uncertainty, the LGM provides a method to guarantee a functional operating domain for the AV. This ensures that the tactical planner receives information with a known confidence level, reducing the risk of hazardous decisions caused by misleading occupancy data. The approach offers a scalable, context-aware alternative to generic occupancy grids, effectively coupling distance metrics with occupancy states to enhance situation understanding in complex urban scenarios.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• situational awareness