Trajectory Planning for Automated Merging Vehicles on Freeway Acceleration Lane

This article presents a trajectory planning framework for an automated merging vehicle on the freeway acceleration lane when it interacts with human-driven mainline vehicles. This is mainly ascribed to the prediction of the average accelerations of the mainline vehicles and a polynomial merging trajectory planning method. The ERT (Extremely Randomized Trees) classifier was utilized to predict the average accelerations of the preceding and following vehicles on the mainline during the merging interaction process based on a naturalistic dataset at the merging zone. The accuracy of the acceleration prediction model for the preceding vehicle on the mainline is 97.7%, and for the following vehicle, it is 96.1%. In addition, an ERT-QP trajectory planning model considering the results of the ERT model and the necessary constraints was proposed based on the Quintic Polynomial (QP) function. The remaining distance in the acceleration lane is considered to guarantee that the lane-crossing behavior occurs before the end of the acceleration lane. The cost of the QP function was formulated based on the stability, duration, terminal speed, and distances between the merging vehicle and the target lane vehicles. At last, a simulation merging scenario on a freeway acceleration lane was established based on the SUMO (Simulation of Urban Mobility) simulation platform and the MATLAB algorithm control platform. The simulation results showed that the ERT-QP trajectory planning model proposed performed better compared to the embedded SL2015 model in SUMO, with a more stable merging trajectory, a shorter merging duration, and a higher merging terminal speed.

I. Introduction

The mixed traffic of automated vehicles and human-driven vehicles (HDVs) is bound to exist for a long time [1]. At the initial stage of the implementation of automated vehicles, most road participants are still HDVs. An automated vehicle needs to correctly perceive and predict the motion states of the surrounding HDVs in order to make appropriate motion plans, achieving direct and safe interaction with them [2], [3].