A Stochastic Hybrid Framework for Driver Behavior Modeling Based on Hierarchical Dirichlet Process

TL;DR

This paper introduces a scalable stochastic hybrid framework using hierarchical Dirichlet processes for modeling driver behavior in vehicle networks, improving performance over baseline methods in real driving data.

Contribution

It presents a novel non-parametric Bayesian approach for joint driver and vehicle behavior modeling, addressing scalability in dense vehicular networks.

Findings

Higher modeling accuracy compared to baseline methods

Effective in real-world driving scenarios

Supports scalable information dissemination in vehicle networks

Abstract

Scalability is one of the major issues for real-world Vehicle-to-Vehicle network realization. To tackle this challenge, a stochastic hybrid modeling framework based on a non-parametric Bayesian inference method, i.e., hierarchical Dirichlet process (HDP), is investigated in this paper. This framework is able to jointly model driver/vehicle behavior through forecasting the vehicle dynamical time-series. This modeling framework could be merged with the notion of model-based information networking, which is recently proposed in the vehicular literature, to overcome the scalability challenges in dense vehicular networks via broadcasting the behavioral models instead of raw information dissemination. This modeling approach has been applied on several scenarios from the realistic Safety Pilot Model Deployment (SPMD) driving data set and the results show a higher performance of this model in comparison with the zero-hold method as the baseline.