자율주행자동차의 실시간 고정밀 위치추정을 위한 파티클필터와 확장칼만필터 융합 위치추정 알고리즘 개발

Abstract

In this paper, we fused an extended Kalman filter and a particle filter to accurately locate autonomous vehicles in real time. While previous researchers have widely used extended Kalman filtering for GPS-based localization of dynamic systems, these solutions have suffered from GPS shadows. Meanwhile, since particle filters do not contain any GPS information, the computational delay from the large number of particles limits the localization precision. We propose a complementary method that compensates the problem of each standalone method: we use the output from the particle filter as a sensed position input to the extended Kalman filter after motion prediction and considering the computational time delay. The combined method is free of GPS shadows and localization errors from the computational delay. We tested this method with an autonomous vehicle system at KOREATECH and showed superior performance compared with an extended Kalman filter and the particle-filter-only localization method.