Identification of dynamic driving styles based on behavioral primitives.

Zheng, X; Kang, W; Ren, Y; Li, X; Xi, J · 2026 · PubMed Central

DOI: 10.1038/s41598-026-38787-y

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the limitations of static driving style classification, which often overlooks the dynamic adaptability of drivers to specific scenarios and temporal fluctuations in behavior. To capture these intrinsic characteristics, the authors propose an unsupervised framework for identifying dynamic driving styles based on "behavioral primitives"—the smallest segments of driving data with clear physical meaning. The study aims to provide a fine-grained, time-series analysis of driver decision-making and vehicle manipulation, offering a more nuanced understanding than traditional holistic or steady-state approaches. The methodology utilizes natural driving data collected from 17 participants in a driving simulator, totaling approximately 4.82 hours. The framework operates in three stages: primitive extraction, model development, and classification. First, a dual-layer hierarchical BMASS (H-BMASS) method segments time-series data into lateral and longitudinal dimensions to identify 2,763 behavioral primitives. These are clustered into five semantic types (e.g., straight medium-speed, turning left) using Latent Dirichlet Allocation (LDA) adapted for temporal features. Second, a dynamic evaluation model quantifies risk using three components: the inherent risk of the current primitive, the risk of transitioning into it, and the risk of transitioning out of it. Feature weights are determined objectively using correlation coefficients and entropy methods, while transition risks incorporate risk differentials and transition probabilities. Finally, an improved Particle Swarm Optimization (PSO) algorithm determines thresholds to classify styles as cautious, average, or aggressive. The results demonstrate that the primitive risk index model effectively distinguishes between the five primitive types, with low overlap in their probability density distributions. The study establishes specific risk indices for each primitive type and constructs a transition risk matrix based on the likelihood and risk differential of moving between states. The dynamic driving style score is calculated via a weighted summation of current and transition risks. The PSO algorithm successfully identifies classification thresholds, allowing for the categorization of the 17 drivers’ dynamic behaviors. The framework captures both the safety of individual actions and the driver’s adaptability to environmental changes, revealing short-term fluctuations that static models miss. The significance of this work lies in its ability to model the temporal continuity of driving behavior, providing a robust tool for analyzing driver decision-making processes. By preserving temporal features and contextual transitions, the framework offers deeper insights into long-term driving behavior and dynamic style variations. These findings support the development of personalized Advanced Driver Assistance Systems (ADAS) and autonomous vehicles (AVs) by enabling more accurate prediction of driver behavior and improved traffic safety analysis.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• human error taxonomy

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: tool software
• Theoretical Contribution: computational model, conceptual framework