신경망 기반 다중모델필터를 이용한 자율주행 자동차의 위치추정

Abstract

The vehicle positioning system is an essential component for developing an autonomous vehicle. Highly accurate and reliable position has to be updated frequently for driving a vehicle autonomously. However, the vehicle positioning system based on a global positioning system (GPS) alone can’t be accurate due to the GPS blockage and a multipath error. In order to overcome this problem, the sensor fusion of the position data from GPS with on-board sensors has been used widely. In general, a single model filter is frequently used for proper positioning of the vehicle, but it is not suitable for various driving conditions. For this reason, a multiple models filter has been studied to apply in a wide-range of driving conditions. The multiple models filter can apply in various conditions because it combines several models which use a model probability, respectively. In this study, we proposed a neural network based on the multiple models filter using the GPS with on-board sensors. The model probability can be obtained by using the off-line learning of the neural network. The proposed multiple models filter were verified by simulation and experiment. The experimental results of the proposed multiple models filter show good performance to provide the accurate vehicle position under the wide-range of driving conditions.