关键词: Markov定位, 特征地图, 高斯核函数, 自动导引车

Abstract: Markov localization algorithm used in the feature map for AGV's global pose estimation might cause the problem that the association between sensor observation and feature data in the map was non-unique,which leaded position estimation for the failure.For this problem,a new calculation method of Markov localization algorithm without data association was proposed.The Gaussian kernel function was used to fit the sparse features of environment into a smooth and dense curve.The observation model of Markov localization algorithm was calculated through comparing the similarity of two dense curves generated by sensor observation and algorithm prediction.The model and algorithm were established,and simulations in two kinds of environment were designed.The simulation in similar environment validated the accuracy of pose estimation,and the simulation in non-similar environment showed the high efficiency of global localization.The validity and superiority of the proposed method were verified by experiments for AGV's pose estimation in a semi-enclosed room.

Key words: Markov localization, feature map, Gaussian kernel function, automated guided vehicle