基于深度学习的激光同步定位与地图构建移动机器人可定位性研究  

Localization of Laser Simultaneous Localization and Mapping Mobile Robot Based on Deep Learning

作　　者：高扬[1] 张传玺 王晨 王书棋 GAO Yang;ZHANG Chuan-xi;WANG Chen;WANG Shu-qi(School of Automobile, Chang’an University, Xi’an 710064, China)

机构地区：[1]长安大学汽车学院,西安710064

出　　处：《科学技术与工程》2021年第32期13774-13780,共7页Science Technology and Engineering

基　　金：国家自然科学基金(61503043);陕西省自然科学基金(2019JLP-07)。

摘　　要：高精度定位是移动机器人执行上层任务的基础,也是直接影响其他任务执行效果的首要问题。可定位性是对定位精度好坏的度量,对可定位性的预估可以使机器人提早避开难以准确定位的区域提高其他任务的成功率。以常用的地图匹配和航迹推算融合实现定位的方法为研究对象,分析了其工作机理,针对定位算法中航迹推算与地图匹配分别设计不同的神经网络模块,形成一种由卷积神经网络(convolutional neural network,CNN)、长短期记忆神经网络(long short-term memory,LSTM)和多层全连接神经网络(multi-layer full connection,MLFC)组成的多模块深度神经网络模型(multi-module deep neural network model,MMN)。所提方法以定位熵表征可定位性,以CNN网络预估地图匹配定位熵,以LSTM网络估计航迹推算定位熵,以MLFC估计前两者融合后的定位熵,从而实现对移动机器人可定位性的估计。仿真和实验结果表明:该模型能够准确估计给定地图上机器人的可定位性,预测熵与实测熵相比误差小于5%。High-precision localization is the basis for mobile robots to perform upper-level tasks,and it is also the primary problem that directly affects the performance of other tasks.Localizability is a measure of positioning accuracy.Estimation of localizability can enable the robot to avoid areas that are difficult to accurately locate and improve the success rate of other tasks.Taking the common methods of map matching and track estimation fusion to realize positioning as the research object,its working mechanism was analyzed.Different neural network modules were designed for track estimation and map matching in positioning algorithm to form a convolution neural network(CNN),long short-term memory(LSTM)neural network,and multi-layer fully connected neural network(MLFC).The localization entropy was used to characterize the localization of the proposed method.Map matching localization entropy was estimated by CNN network,dead-reckoning localization entropy was estimated by LSTM network,and localization entropy after the fusion of the first two was estimated by MLFC,so as to realize the estimation of localization of mobile robot.The simulation and experimental results show that the proposed method can offer a direct and accurate prediction of the localizability of robots on a given map,the error between the predicted entropy and the measured entropy is less than 5%.

关 键 词：可定位性 卷积神经网络(CNN) 长短期记忆(LSTM)神经网络 航迹推算 地图匹配 熵值 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]