Intelligent Control of Merging Car-following and Lane-Changing Behavior
Farzam Tajdari, Amin Rezasoltani
TL;DR
The paper addresses transient merging by introducing a perception-index that decomposes follower-vehicle behavior into anticipation, perception, preparation, and relaxation. It develops a perception-based ANFIS fuzzy controller trained on NGSim driver data to drive the follower during lane-change events, using inputs derived from lateral velocity, jerk-like perception signals, and inter-vehicle spacing. Through closed-loop plant simulations and comparisons with real drivers and a prior controller, the approach yields smoother, safer, and more comfortable trajectories while maintaining distances aligned with Pipe's law, potentially reducing traffic delays and motion sickness. The work demonstrates the viability of incorporating measurable human perception into autonomous driving controllers and outlines future stability analyses and real-world testing.
Abstract
Recent research has paid little attention to complex driving behaviors, namely merging car-following and lane-changing behavior, and how lane-changing affects algorithms designed to model and control a car-following vehicle. During the merging behavior, the Follower Vehicle (FV) might significantly diverge from typical car-following models. Thus, this paper aims to control the FV witnessing lane-changing behavior based on anticipation, perception, preparation, and relaxation states defined by a novel measurable human perception index. Data from human drivers are utilized to create a perception-based fuzzy controller for the behavior vehicle's route guidance, taking into account the opacity of human driving judgments. We illustrate the efficacy of the established technique using simulated trials and data from actual drivers, focusing on the benefits of the increased comfort, safety, and uniformity of traffic flow and the decreased of wait time and motion sickness this brings about.