GNSS based Lane Keeping Assist System via Model Predictive Control 2019-01-0685

Recently, the field of autonomous driving has been dramatically expanding, and some of key technologies like Lane Keeping Assist (LKA) system have begun to be applied to mass production vehicles. In general, mass-produced LKA systems use a lane detection camera as a means of keeping a lane. One of the common limitations of camera based LKA systems is that the lane keeping performance will significantly decreases when the camera cannot detect lane markings due to various reasons such as snow coverage and sunlight. To overcome this limitation, we have developed Global Navigation Satellite Systems (GNSS) based LKA systems, which are not affected by surrounding environment such as weather and lighting. Our LKA systems use centimeter level augmentation service and high definition maps, whereby the LKA systems are able to accurately estimate self-position. This feature enables the LKA systems to potentially show higher lane keeping performance than camera based LKA systems even when lane markings are undetectable. In our previous study, we proposed a GNSS based LKA system in which target steering wheel angle was calculated by means of PID controller based on a look-ahead model. Although there were a few problems such as oscillation of steering, the proposed system enabled a real vehicle to keep a lane even under conditions in which camera based LKA systems would not probably work well. In this paper, to aim at improving lane keeping performance, we proposed a GNSS based LKA system which calculates target steering wheel angle via Model Predictive Control (MPC). We then validated the lane keeping performance of the LKA system using MPC in both of simulation and real vehicle tests.