Learning Collision Risk from Naturalistic Driving with Generalised Surrogate Safety Measures

2025-05-19

AI Summary

Collision warning systems in urban interactive traffic suffer from delayed alerts, poor generalizability, and reliance on manually annotated risk labels. Method: This paper proposes Generalized Surrogate Safety Measure (GSSM), an end-to-end risk learning framework that requires no collision or risk labels. It introduces a novel context-adaptive risk scoring mechanism grounded in Extreme Value Theory (EVT), where neural networks model the conditional distribution of multi-directional inter-vehicle distances, integrating heterogeneous naturalistic driving data—including weather, illumination, and kinematic features—to autonomously learn normal interaction patterns and quantify deviations from safe behavioral norms. Contribution/Results: Evaluated on 4,875 real-world near-crash/crash events reconstructed from the SHRP2 Naturalistic Driving Study, GSSM achieves an AUPRC of 0.9 and provides an average lead time of 2.6 seconds. It consistently outperforms existing baselines across diverse scenarios—including rear-end, lane-change, and intersection conflicts—demonstrating strong generalizability and suitability for real-time deployment.

Technology Category

• Natural Language Processing: Safety and Robustness
• Multiagent Systems: Adversarial Agents
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Accurate and timely alerts for drivers or automated systems to unfolding collisions remains a challenge in road safety, particularly in highly interactive urban traffic. Existing approaches require labour-intensive annotation of sparse risk, struggle to consider varying interaction context, or are useful only in the scenarios they are designed for. To address these limits, this study introduces the generalised surrogate safety measure (GSSM), a new approach that learns exclusively from naturalistic driving without crash or risk labels. GSSM captures the patterns of normal driving and estimates the extent to which a traffic interaction deviates from the norm towards unsafe extreme. Utilising neural networks, normal interactions are characterised by context-conditioned distributions of multi-directional spacing between road users. In the same interaction context, a spacing closer than normal entails higher risk of potential collision. Then a context-adaptive risk score and its associated probability can be calculated based on the theory of extreme values. Any measurable factors, such as motion kinematics, weather, lighting, can serve as part of the context, allowing for diverse coverage of safety-critical interactions. Multiple public driving datasets are used to train GSSMs, which are tested with 4,875 real-world crashes and near-crashes reconstructed from the SHRP2 NDS. A vanilla GSSM using only instantaneous states achieves AUPRC of 0.9 and secures a median time advance of 2.6 seconds to prevent potential collisions. Additional data and contextual factors provide further performance gains. Across various interaction types such as rear-end, merging, and crossing, the accuracy and timeliness of GSSM consistently outperforms existing baselines. GSSM therefore establishes a scalable, context-aware, and generalisable foundation to proactively quantify collision risk in traffic interactions.

Problem

• Develops a method to predict collision risk without crash data
• Uses naturalistic driving patterns to assess safety deviations
• Improves accuracy and timeliness of collision warnings in diverse scenarios

Innovation

• Uses neural networks to analyze normal driving patterns
• Calculates context-adaptive risk scores with extreme value theory
• Learns from naturalistic driving without crash labels