ST-GraphNet: A Spatio-Temporal Graph Neural Network for Understanding and Predicting Automated Vehicle Crash Severity

2025-06-09

AI Summary

Accurately predicting collision severity for autonomous vehicles (AVs) remains challenging due to the complex interplay of spatiotemporal dynamics, automation levels, and unstructured incident narratives. Method: This paper proposes a dual-scale graph modeling framework: (i) a fine-grained event graph capturing individual collisions, and (ii) an H3 hexagonal regional aggregation graph encoding spatial contiguity and density. The framework fuses multimodal features—including spatiotemporal coordinates, SAE automation levels, and semantic embeddings from accident reports—within a novel Dynamic Spatio-Temporal Graph Convolutional Network (DSTGCN), which incorporates adaptive spatial aggregation and dynamic message passing. Contribution/Results: Evaluated on 2,352 real-world AV collision records, the model achieves 97.74% accuracy—substantially outperforming its single-scale fine-grained counterpart (64.7%). It enables precise identification of urban high-risk zones and delivers an interpretable, production-ready paradigm for intelligent transportation risk assessment and infrastructure planning.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Security and Privacy: Large-scale security measurements

Abstract

Understanding the spatial and temporal dynamics of automated vehicle (AV) crash severity is critical for advancing urban mobility safety and infrastructure planning. In this work, we introduce ST-GraphNet, a spatio-temporal graph neural network framework designed to model and predict AV crash severity by using both fine-grained and region-aggregated spatial graphs. Using a balanced dataset of 2,352 real-world AV-related crash reports from Texas (2024), including geospatial coordinates, crash timestamps, SAE automation levels, and narrative descriptions, we construct two complementary graph representations: (1) a fine-grained graph with individual crash events as nodes, where edges are defined via spatio-temporal proximity; and (2) a coarse-grained graph where crashes are aggregated into Hexagonal Hierarchical Spatial Indexing (H3)-based spatial cells, connected through hexagonal adjacency. Each node in the graph is enriched with multimodal data, including semantic, spatial, and temporal attributes, including textual embeddings from crash narratives using a pretrained Sentence-BERT model. We evaluate various graph neural network (GNN) architectures, such as Graph Convolutional Networks (GCN), Graph Attention Networks (GAT), and Dynamic Spatio-Temporal GCN (DSTGCN), to classify crash severity and predict high-risk regions. Our proposed ST-GraphNet, which utilizes a DSTGCN backbone on the coarse-grained H3 graph, achieves a test accuracy of 97.74%, substantially outperforming the best fine-grained model (64.7% test accuracy). These findings highlight the effectiveness of spatial aggregation, dynamic message passing, and multi-modal feature integration in capturing the complex spatio-temporal patterns underlying AV crash severity.

Problem

• Predict AV crash severity using spatio-temporal graph neural networks
• Model fine-grained and region-aggregated spatial crash dynamics
• Integrate multimodal data to classify severity and high-risk regions

Innovation

• Spatio-temporal graph neural network for AV crash prediction
• Hexagonal Hierarchical Spatial Indexing for spatial aggregation
• Dynamic Spatio-Temporal GCN for high accuracy