Design and Evaluation of Lane Change Decision for Automated Driving Vehicle Using Stochastic Predictive Control

Abstract

The traffic accidents caused by human error, such as distraction, drowsiness, or mistakes, account for 94 percent of all traffic accidents over last decades. Since safe driving is a goal of road-traffic-vehicle environments, major automakers develop driver assistance and active safety system as key technologies since the early 1990s. For instance, parking assistance system (PAS), lane keeping assistance system (LKAS), smart cruise control (SCC), and automated emergency braking (AEB) already have been developed and commercialized by major automakers. Moreover, in recent years, an interest of automotive industry is changed from the development of active safety to that of automated driving system capable of sensing surrounding environments and driving itself. Google starts to operate an automated driving vehicle on real roads using a pre-measured precise map and environmental information from a laser scanner. The electric vehicle mass production company, Tesla, has applied the autopilot mode to Model S vehicle. BMW has succeeded driving a vehicle autonomously in real traffic from Munich to Ingolstadt in German with robustness, and safety. Mercedes-Benz developed Intelligent Drive system and followed the route with fully autonomous driving from Mannheim to Pforzheim in German. These trends of automated driving vehicle development promote not only safety of passengers but also convenience. However, the current state-of-the-art in automated driving technology demands sophisticated decision to manage various and complex traffic circumstances. This dissertation focused on the development of an automated driving control algorithm which can determine appropriate vehicle motion to deal with complex situations such as merging roads in the highway. In order to enhance decision to drive a vehicle safely, the potential behaviors of surrounding vehicles, the current and predicted states with sensor uncertainties of surrounding vehicles should be considered. Based on the prediction of surrounding vehicles, the lane change or keeping mode is determined by risk monitoring. A safe driving envelope which indicates the safe drivable area is also defined in consideration of prediction states of other traffics. The subject vehicle plans lane change or keeping motion such as accelerating, overtaking, and braking. To obtain desired control inputs which have been planned in advance, the predictive control problem is formulated. However, for automated driving vehicles in dynamic road environments, the uncertainty has to be considered from states of vehicle dynamic model and actual sensing values. For this reason, Stochastic Model Predictive Control with adaptive uncertainty propagation is developed to improve performance. The performance of the proposed algorithm is validated via computer simulations and vehicle tests. Automated driving with the proposed algorithm shows smooth and safe driving behavior in various road traffic situations, such as lane keeping with preceding vehicle following, lane change in a multi-vehicle environment. The effectiveness of the proposed automated driving algorithm is evaluated via vehicle tests. Test results show the robust performance on a motorway scenario.