摘要
在概率机器人学中,基于距离传感器的概率测量模型被提出并被广泛地应用到移动机器人的定位及建图过程中。在复杂多变的外界环境下,为了得到更符合实际情况、更精确的概率测量模型,提出了一种在线学习的方法:根据实时冗余的传感器数据,按真实障碍物距离进行分类得到相应的数据集,并采用期望最大化(EM)算法从各数据集中学习相应的模型内参,从而对概率测量模型持续更新与校正。实验结果表明,采用基于在线学习的概率测量模型,能得到鲁棒性更强、精度更高的定位结果。
In probabilistic robots,probabilistic measurement model of range sensors has been proposed and ap- plied in robot localization and mapping widely.Under the complex and volatile environment, in order to obtain more realistic and more accurate model,a method based on online learning was proposed.According to real ranges to ob- stacles, the real-time and redundancy sensor datasets were clustered into several datasets,and then the model intrin- sic parameters have been learned from these datasets by EM algorithm to realize the goal of updating and calibrating the measurement model continuously.The results suggested that it can obtain more precise and robust localization re- sults by applying online learning algorithm to probabilistic measurement model.
作者
何叶
熊根良
HE Ye;XIONG Genliang(School of Mechatronies Engineering,Nanchang University,Nanchang 330031,China)
出处
《南昌大学学报（工科版）》
CAS
2018年第3期286-291,共6页
Journal of Nanchang University(Engineering & Technology)
基金
国家自然科学基金资助项目(61263045
61763030)
江西省科技支撑项目(20112BBE50017)
关键词
在线学习
EM算法
概率测量模型
模型内参
online learning
EM algorithm
probabilistic measurement model
intrinsic parameters
