Simultaneous Estimation of Vehicle Position and Data Delays using Gaussian Process based Moving Horizon Estimation

Abstract

Automobiles or robots with advanced autonomous systems are equipped with multiple types of sensors to overcome different weather and geographical conditions. These sensors generally have various data delays and sampling rates. Additionally, the communication delays or time synchronization errors between the onboard computers significantly affect the robustness and accuracy of localization for autonomous vehicles. In this paper, the simultaneous estimation of vehicle position and sensor delays using a Gaussian process based moving horizon estimation (GP-MHE) is presented. The GP-MHE can estimate the unknown delays of multiple sensors with the resolution less than that of GP-MHE sampling rate. The localization performance of GP-MHE was confirmed using full-vehicle simulator, then evaluated in a real vehicle experiment on a highway scenario. Experimental result verified the sufficient localization accuracy of sub 0.3m using data that had irregular sampling rate and delay of more than 150ms. The proposed algorithm extends the capability of integrating various data with large unknown delays for vehicles, robots, drones and remote autonomy.