A Survey on Data-Driven Modeling of Human Drivers’ Lane-Changing Decisions

Linxuan Huang, Dong-Fan Xie, Li Li, Fellow, IEEE, Zhengbing He, Senior Member, IEEE · 2025

Read A Survey on Data-Driven Modeling of Human Drivers’ Lane-Changing Decisions by Linxuan Huang, Dong-Fan Xie, Li Li, Fellow, IEEE, Zhengbing He, Senior Member, IEEE (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-06680.txt.

Introduction

Linxuan Huang and co-authors survey how data-taught models call a driver's lane change, in a survey from 2025. Lane change is a small act with big wake: safety and flow both ride it. Old hand-built rules call it in stiff forms that miss how folk split and mix on real roads.

Taught models read rich road banks and learn the hid marks of a choice. The survey walks the whole build: banks and their clean-up, in and out marks, aims, frames, and grade ways. It ends with the gaps: crash-safe calls, doubt kept in hand, and one frame that fits smart and self-drive fleets. The field, they say, has the data; it owes the care.

Catalog record

CREATORLinxuan Huang, Dong-Fan Xie, Li Li, Fellow, IEEE, Zhengbing He, Senior Member, IEEE
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE65 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-06680

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Survey on Data-Driven Modeling of Human Drivers’ Lane-Changing Decisions Linxuan Huang, Dong-Fan Xie, Li Li, Fellow, IEEE, Zhengbing He, Senior Member, IEEE Abstract Lane-changing (LC) behavior, a critical yet complex driving maneuver, significantly influences driving safety and traffic dynamics. Traditional analytical LC decision (LCD) models, while effective in specific environments, often oversimplify behavioral heterogeneity and complex interactions, limiting their capacity to capture real LCD. Data-driven approaches address these gaps by leveraging rich empirical data and machine learning to decode latent decision-making patterns, enabling adaptive LCD modeling in dynamic environments. In light of the rapid development of artificial intelligence and the demand for data-driven models oriented towards connected vehicles and autonomous vehicles, this paper presents a comprehensive survey of data-driven LCD models, with a particular focus on human drivers’ LC decision-making. It systematically reviews the modeling framework, covering data sources and preprocessing, model inputs and outputs, objectives, structures, and validation methods. This survey further discusses the opportunities and challenges faced by data-driven LCD models, including driving safety, uncertainty, as well as the integration and improvement of technical frameworks. I Introduction Lane-changing (LC), as a fundamental driving maneuver, exerts a profound influence on road safety [ 1 , 2 , 3 , 4 ] and traffic flow dynamics [ 5 , 6 , 7 , 8 ] . Compared to car-following (CF) behavior, LC behavior entails higher collision risks due to its dependency on holistic evaluations of traffic conditions in both the original and target lanes, requiring drivers to navigate multi-criteria decision-making processes. More specifically, safe LC execution necessitates gaps in the target lane to satisfy collision-avoidance criteria. Drivers must continuously monitor the real-time states of surrounding vehicles (e.g., velocity, acceleration) and adjust their LC maneuvers in response to unexpected behavioral changes (e.g., sudden deceleration, lane encroachment). Human drivers’ irrational decision-making (e.g., sudden risk-preference shifts) in dynamic environments pose challenges to traditional LC models based on hypothesis of rational man. This complex decision-making process significantly impacts traffic safety and traffic efficiency. Consequently, it is essential to deeply explore the mechanisms underlying LC decisions and develop LC models capable of accurately characterizing the decision-making process. This effort will provide critical support for trustworthy traffic simulations, dynamic traffic management, and LC decision-making of autonomous vehicles (AVs). Generally speaking, LC process comprises two components: LC decision (LCD) and LC implementation (LCI) [ 9 ] . LCD, as the critical cognitive component that determines the feasibility and safety of a maneuver, has received predominant attention in lane-changing modeling studies. The modeling of LCD generally focuses on three fundamental research questions: (1) What are the key determinants affecting drivers’ LCD under dynamic traffic conditions? (2) How can we establish a generalized computational framework that accurately captures the nonlinear relationships between influencing factors and LCD behaviors? (3) What methodological innovations are required to develop LCD models that demonstrate robust generalization capabilities across diverse driving scenarios? Analytical LCD models, grounded in mathematical formulations and probabilistic theories, encompass both rule-based reactive models [ 10 , 11 , 12 , 13 ] and game-theoretic interactive models [ 14 , 15 , 16 ] . These models, characterized by well-defined physical principles and structured modeling procedures, are typically categorized as analytical approaches. Once calibrated, these models demonstrate proficiency in executing LC tasks within constrained environments, such as specific roadways. However, their focus on abstracting and simplifying driving behavior patterns makes it difficult to explicitly represent latent behavioral characteristics, such as driving style heterogeneity and social collaboration tendencies. This limitation significantly restricts their generalizability across diverse traffic scenarios and varying traffic conditions over time. In contrast, data-driven models leverage real driver behavior data to uncover complex behavioral patterns and latent features through learning-based approaches. This approach effectively establishes a mapping from diverse inputs to LC behavior, thereby inherently excelling in achieving human-like decision-making capabilities. Such models facilitate the transition from group behavioral norms to individualized characteristics, enabling the implementation of dynamic personalized LC strategies and adaptability to complex interactive scenarios in open-road environments. Furthermore, advancements in data acquisition technologies and distributed computing frameworks have provided robust foundations for data-driven LCD modeling by enabling the collection and utilization of large-scale, high-precision vehicle trajectory data. Existing data-driven LC models can be broadly categorized into the following two groups. Human behavior-oriented models, which are designed to understand and describe human drivers’ LC behavior; AV-oriented models, which aim at enhancing AV performance. These two groups of models differ fundamentally in objectives, methodologies, and applications. • Human behavior-oriented models explore the “why” behind lane changes, capturing subjective factors (e.g., aggressive/conservative styles) and stochastic interactions (e.g., unspoken cooperation) [ 9 ] . They rely on probabilistic models and supervised learning to predict LC behaviors, ultimately improving the applications such as human-centric warnings in advanced driver assistance systems (ADAS). • AV-oriented models solve the “how” of lane changes, converting sensor data into optimal actions via various modeling frameworks (e.g., reinforcement learning rewards, game theory). They enforce objective-driven decisions (e.g., safety guarantees, traffic efficiency) and integrate with control systems for machine-executable maneuvers [ 17 , 18 , 19 , 20 ] . Despite their differences, the two perspectives are interconnected: insights from human behavior models can inform AV algorithms to improve human-like performance, while data from AVs can refine behavior models by offering a broader range of observations and more granular behavioral patterns, especially in complex traffic scenarios. This design emphasizes robustness and adaptability to ensure the safe integration of AVs into traffic systems. This survey primarily focuses on reviewing data-driven modeling of human drivers’ LCD, covering their methodologies, structural frameworks, existing challenges, and emerging trends. While partially covering LCD models for AVs, the primary aim remains to evaluate their system-level impacts on traffic flow dynamics, such as traffic simulations, rather than delving into vehicle-specific control strategies (e.g., perception systems, real-time motion planning). By maintaining this scope, we aim to clarify how to develop data-driven models that capture LCD behaviors, while also acknowledging the increasing interplay between AV technologies and traffic flow modeling. Given the vastness of the topic, the following is excluded from the scope of this study: (1) Driver physiological/psychological states (e.g., distracted driving, health abnormalities, and other confounding factors); (2) Studies that focus solely on vehicle motion without integrating data-driven insights into individual driver behavior; (3) Moral dilemma scenarios (e.g., trolley dilemmas), which are excluded due to their incompatibility with microscopic driving models. An increasing number of researchers are acknowledging the benefits of utilizing data-driven methods in understanding and modeling LCD behaviors. These methods utilize substantial trajectory data to capture complex driving behaviors and enhance models through machine learning (ML) and statistical analysis techniques. Despite the extensive literature, our objective is to shed light on the following issues that have not received sufficient attention and provide a concise technical framework. • The application of data-driven algorithms in LCD has been extensively reviewed. These algorithms leverage historical and real-time traffic data to model human drivers’ LCD effectively. The integration of data-driven approaches into LCD modeling represents a potential paradigm shift in the field. • This survey provides a comprehensive and thorough review of existing data-driven LCD models. It summarizes the key characteristics of these models, including their data sources, data processing methods, inputs and outputs, structure, objectives, and

Read the complete text