Counterfactual Occlusion-Aware Learning via Visibility Intervention for LiDAR Anomaly Detection

LiDAR point cloud anomaly detection is critical for autonomous system safety, yet most existing methods rely only on visible measurements, overlooking occlusion as a structured consequence of the LiDAR sensing process. We argue that anomalies are characterized not only by what is observed, but also by the spatial voids they create, which alter occlusion patterns and volumetric visibility. We propose Counterfactual Occlusion-Visibility Anomaly Learning (COVAL), a framework that intervenes on volumetric visibility during training. Using physics-conformed synthetic anomaly construction, COVAL generates paired factual and counterfactual observations with identical scene geometry but different occlusion patterns. Then, we introduce two complementary objectives: Visibility-Variant Counterfactual Reconstruction, which models occlusion-induced missing regions, and Visibility-Invariant Counterfactual Consistency, which enforces stable representations across visibility changes. Together, these objectives isolate anomaly-induced structural missingness and in turn refine representation of normal scenes, thus improving anomaly sensitivity at test time. Experiments on standard LiDAR anomaly segmentation benchmarks show that COVAL achieves state-of-the-art performance.

Autonomous vehicles use LiDAR sensors to perceive the 3D world, but they must also detect unexpected objects or structures that were not seen during training. Most existing methods focus mainly on the points directly observed by LiDAR, while overlooking an important clue: the missing regions created when objects block the sensor’s view. These occluded regions can reveal whether a scene follows normal structure or contains an abnormal object. We propose COVAL, a training framework that creates synthetic anomalies in normal LiDAR scenes and constructs paired views with different visibility patterns. By comparing what changes and what remains stable across these views, the model learns to distinguish normal occlusion from anomaly-induced missing regions. This counterfactual visibility information is used only during training. At test time, the model still takes a single LiDAR scan as input. Experiments show that COVAL improves LiDAR anomaly segmentation, which can support safer perception in autonomous systems.