A Hybrid Queuing Model for Coordinated Vehicle Platooning on Mixed-Autonomy Highways: Training and Validation

TL;DR

This paper introduces a hybrid queuing model and real-time training algorithm for predicting vehicle counts and optimizing platooning on mixed-autonomy highways, enhancing system efficiency and decision-making.

Contribution

It presents a novel hybrid queuing model with an online training algorithm for real-time highway parameter estimation in mixed-autonomy traffic.

Findings

Prediction error ~15% in stationary settings

Prediction error ~25% in non-stationary settings

Trained model achieves near-optimal platoon headway regulation

Abstract

Platooning of connected and autonomous vehicles (CAVs) is an emerging technology with a strong potential for throughput improvement and fuel reduction. Adequate macroscopic models are critical for system-level efficiency and reliability of platooning. In this paper, we consider a hybrid queuing model for a mixed-autonomy highway section and develop an easy-to-use training algorithm. The model predicts CAV and non-CAV counts according to the traffic demand as well as key parameters of the highway section. The training algorithm learns the highway parameters from observed data in real time. We test the model and the algorithm in Simulation of Urban Mobility (SUMO) and show that the prediction error is around 15% in a stationary setting and around 25% in a non-stationary setting. We also show that the trained model leads to a platoon headway regulation policy very close to the simulated optimum. The proposed model and algorithm can directly support model-predictive decision-making for platooning in mixed autonomy.