基于视觉和轮速计紧耦合的轮式机器人定位算法研究  

Tightly-Coupled Vision and Wheel Odometer Localization for Ground Vehicle

作　　者：朱道俊 徐湛楠 马婷婷[1] 曹平国 宋全军[1] ZHU Daojun;XU Zhannan;MA Tingting;CAO Pingguo;SONG Quanjun(Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei Anhui 230031,China;Scinece Island Branch,Graduate School of USTC,Hefei Anhui 230026,China)

机构地区：[1]中国科学院合肥物质科学研究院,安徽合肥230031 [2]中国科学技术大学研究生院科学岛分院,安徽合肥230026

出　　处：《传感技术学报》2021年第7期896-903,共8页Chinese Journal of Sensors and Actuators

基　　金：安徽省重点研究与开发计划项目(201904d07020007,202004a07020051)。

摘　　要：轮式机器人在未知环境下的定位是机器人学中一个经过广泛研究但仍然需要进一步提高的问题。VINS-Mono是目前效果最好的算法之一,然而其应用在轮式机器人上会因退化问题导致定位精度下降。本文通过充分利用轮速计与视觉系统之间的耦合关系,提出了一种基于视觉和轮速计紧耦合的轮式机器人定位系统估计器。针对低频视觉信息和高频轮速计信息的融合问题,文中推导了轮速计预积分测量和误差传播过程。针对单目视觉无尺度初始化和VINS-Mono在轮式机器人上的初始化退化问题,文中提出了一种联合视觉和轮速计信息的快速精确系统初始化流程。通过在代表性室内及室外环境的实际验证,本文所提出的方法不但比VINS-Mono具有更好的定位精度,而且其计算复杂度仅为VINS-Mono的约十分之一。The Localization of ground vehicle in unknown environments is a problem in robotics that has undergone extensive research but still needs further improvement. VINS-Mono is currently one of the best effective algorithms, but its application on wheeled robots will cause degradation of localization accuracy due to degradation. By making full use of the coupling relationship between the wheel odometer and the vision system, this paper proposes a wheeled robot localization system estimator based on the tight coupling of vision and wheel odometer. Aiming at the fusion of low-frequency visual information and high-frequency wheel speedometer information, the pre-integration measurement and error propagation process of wheel speedometer are deduced in this paper. Aiming at the scale-free initialization of monocular vision and the initialization degradation of VINS-Mono on wheeled robots, this paper proposes a fast and accurate system initialization process that combines vision and wheel odometer information. Through actual verification in representative indoor and outdoor environments, the method proposed in this paper not only has better localization accuracy than VINS-Mono, but its computational complexity is only about one-tenth of VINS-Mono.

关 键 词：同时建图与定位 机器人定位 传感器融合 轮式机器人 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]