[Paper Review] Modeling and Prediction of Human Driver Behavior: A Survey.

Kyle Brown, Katherine Driggs-Campbell+1 more|arXiv (Cornell University)|Jun 15, 2020

Autonomous Vehicle Technology and Safety18 citations

TL;DR

This paper presents a comprehensive survey and taxonomy of 200 driver behavior models, using a partially observable stochastic game framework to unify and compare models across core tasks like state and intention estimation, trait modeling, and motion prediction. The key contribution is a structured classification of models based on their tasks and methodological attributes, enabling clearer understanding and comparison in autonomous driving research.

ABSTRACT

We present a review and taxonomy of 200 models from the literature on driver behavior modeling. We begin by introducing a mathematical formulation based on the partially observable stochastic game, which serves as a common framework for comparing and contrasting different driver models. Our taxonomy is constructed around the core modeling tasks of state estimation, intention estimation, trait estimation, and motion prediction, and also discusses the auxiliary tasks of risk estimation, anomaly detection, behavior imitation and microscopic traffic simulation. Existing driver models are categorized based on the specific tasks they address and key attributes of their approach.

Motivation & Objective

• To address the growing complexity and fragmentation in driver behavior modeling by providing a unified framework for comparison.
• To identify and categorize the core modeling tasks in driver behavior research, including state estimation, intention inference, trait modeling, and motion prediction.
• To analyze the methodological diversity across 200 existing models and classify them based on task focus and key attributes.
• To support future research by clarifying the roles of auxiliary tasks such as risk estimation, anomaly detection, behavior imitation, and microscopic traffic simulation.
• To establish a foundation for systematic development and evaluation of driver behavior models in autonomous vehicle systems.

Proposed method

• The authors develop a mathematical formulation based on the partially observable stochastic game (POSG) to serve as a common theoretical framework for modeling driver behavior.
• They define a taxonomy centered on four core modeling tasks: state estimation, intention estimation, trait estimation, and motion prediction.
• Models are classified according to their primary task focus and key methodological attributes, such as data sources, inference techniques, and modeling assumptions.
• The approach integrates auxiliary tasks like risk estimation, anomaly detection, behavior imitation, and microscopic traffic simulation into the overall framework.
• The taxonomy enables systematic comparison and contrast of models across different research directions and applications.

Experimental results

Research questions

• RQ1 How can a unified framework be established to compare diverse driver behavior models in the literature?
• RQ2 What are the primary modeling tasks that define the scope and function of driver behavior models?
• RQ3 How do different models vary in their approach to state, intention, and trait estimation?
• RQ4 What role do auxiliary tasks such as risk estimation and anomaly detection play in comprehensive driver modeling?
• RQ5 How can the taxonomy support the development of more accurate and interoperable models for autonomous driving systems?

Key findings

• The partially observable stochastic game framework successfully unifies diverse driver behavior models under a common mathematical structure.
• State estimation, intention estimation, trait estimation, and motion prediction are the four dominant core modeling tasks identified in the literature.
• A significant portion of models focus on motion prediction, often using probabilistic or deep learning methods to forecast trajectories.
• Auxiliary tasks such as risk estimation and anomaly detection are increasingly integrated into models, especially in safety-critical applications.
• The taxonomy enables clear differentiation between models based on task focus, data modality, and inference technique, enhancing model selection and system design.
• The survey reveals methodological diversity, with no single dominant approach, highlighting the need for standardized evaluation benchmarks.