A Graph-based Framework for Coverage Analysis in Autonomous Driving

TL;DR

The paper introduces a graph-based framework for coverage analysis in autonomous driving by modeling traffic scenes as hierarchical graphs that merge map topology with actor relations. It presents two complementary coverage analyses: a subgraph isomorphism approach against manually defined archetypes for interpretable coverage, and a graph embedding approach using Graph Isomorphism Networks with Edge features (GINE) trained via self-supervised contrastive learning to enable similarity-based coverage and anomaly detection. Validated on real-world Argoverse 2.0 data and CARLA simulations, the results reveal meaningful structural differences and coverage gaps, with 62% overall archetype coverage and distinct distributional disparities between datasets. The framework scales to arbitrary actor counts and reduces the need for scenario-specific handling, with future work focusing on temporal extension and automatic archetype extraction; code is publicly available.

Abstract

Coverage analysis is essential for validating the safety of autonomous driving systems, yet existing approaches typically assess coverage factors individually or in limited combinations, struggling to capture the complex interactions inherent in traffic scenes. This paper proposes a graph-based framework for coverage analysis that represents traffic scenes as hierarchical graphs, combining map topology with actor relationships. The framework introduces a two-phase graph construction algorithm that systematically captures spatial relationships between traffic participants, including leading, following, neighboring, and opposing configurations. Two complementary coverage analysis methods are presented. First, a sub-graph isomorphism approach matches traffic scenes against a set of manually defined archetype graphs representing common driving scenarios. Second, a graph embedding approach utilizes Graph Isomorphism Networks with Edge features (GINE) trained via self-supervised contrastive learning to project traffic scenes into a vector space, enabling similarity-based coverage assessment. The framework is validated on both real-world data from the Argoverse 2.0 dataset and synthetic data from the CARLA simulator. The subgraph isomorphism method is used to calculate node coverage percentages using predefined archetypes, while the embedding approach reveals meaningful structure in the latent space suitable for clustering and anomaly detection. The proposed approach offers significant advantages over traditional methods by scaling efficiently to diverse traffic scenarios without requiring scenario-specific handling, and by naturally accommodating varying numbers of actors in a scene.

Figures (14)

• Figure 1: The map graph (left) serves as the foundational structure for determining actor relations, encoding spatial relationships between lanes. The traffic scene (right) shows actors and their relationships at a specific timestep. Note that actors can be disconnected if they are far enough apart, as the graph construction algorithm only creates edges between actors within the specified distance thresholds.
• Figure 2: Comparison of the actor graph after relation discovery and after hierarchical selection with redundancy prevention. The discovery graph contains all relations found within the distance limits, while the final graph removes redundant edges that can be represented through existing paths, significantly reducing the number of edges while preserving connectivity.
• Figure 3: Model architecture for the Graph Isomorphism Network with Edge features (GINE).
• Figure 4: Overview of CARLA maps used in the simulation study: Town01, Town02, Town03, Town04, Town05, and Town07. These maps provide diverse urban driving environments with varying road layouts, intersections, and traffic patterns.
• Figure 5: Coverage barcharts for the manually defined coverage scenarios for CARLA and Argoverse.