LARAD: Layout-Aware Road Anomaly Detection via Spatial-Logic Reasoning

2026-07-14

AI Summary

This work addresses the limitations of existing road anomaly detection methods, which overly rely on texture novelty while neglecting spatial-logical context, often leading to missed detections or false alarms for logically anomalous objects and requiring cascaded large models that incur high latency. To overcome these issues, the authors propose LARAD, a novel approach that shifts the anomaly detection paradigm from appearance matching to spatial-logical reasoning. LARAD introduces a pioneering Spatial-Logical Violation Synthesis (SLVS) mechanism to generate training samples that are texturally consistent yet logically anomalous, and incorporates a lightweight out-of-distribution (OoD)-guided attention module to enhance a closed-set semantic segmentation network. Operating within a single-model architecture without cascading, LARAD efficiently detects logical anomalies with low inference latency while achieving state-of-the-art performance and improved robustness in road anomaly detection.

Abstract

Accurate open-world obstacle detection is critical for autonomous driving. Current anomaly segmentation methods suffer from a fundamental blind spot: they over-rely on texture novelty to identify out-of-distribution (OoD) objects while ignoring contextual spatial logic. Furthermore, mitigating the resulting false positives often requires cascading massive vision models, introducing unacceptable inference latency. To address these issues, we propose Layout-Aware Road Anomaly Detection (LARAD), shifting the paradigm from appearance matching to spatial-logic reasoning. First, we introduce the Spatial-Logic Violation Synthesis (SLVS) pipeline, which generates training samples that are texture-consistent yet spatially invalid, forcing the model to learn contextual violations. Second, we augment a standard closed-set segmentation network with a lightweight, OoD-guided attention branch. Extensive experiments demonstrate that LARAD significantly enhances robustness against logical anomalies and establishes a new state-of-the-art, all while retaining the high efficiency of a single-model architecture.

Problem

• anomaly detection
• spatial-logic reasoning
• out-of-distribution
• autonomous driving
• false positives

Innovation

• spatial-logic reasoning
• anomaly detection
• out-of-distribution
• layout-aware
• lightweight attention