大规模复杂场景下室内服务机器人导航的研究

发布时间：2018-02-27 09:55

本文关键词： 四叉树地图表示局部路径规划里程计标定人机导航服务机器人　出处：《中国科学技术大学》2017年博士论文　论文类型：学位论文

【摘要】：机器人导航是机器人领域的一项基本研究,其重要意义在于:在所处环境中自主移动的是很多种类机器人能够完成其它复杂任务的前提。近几十年来,随着机器人技术和人工智能技术的不断发展,以及整个社会对服务机器人日益增长的使用需求,学术界和工业界都投入大量的资源对机器人导航技术进行了深入研究和应用探索,使得室内机器人的导航技术日趋成熟。室内机器人的导航技术大体可分为三个发展阶段:工业机器人阶段,家庭服务机器人阶段和公共服务机器人阶段。公共服务领域场景相比工厂场景和普通家庭场景更加复杂,这给现阶段以适应普通家庭环境为导向的室内机器人导航技术带来新的问题和挑战,这也是本文研究的大背景。目前主流的室内机器人导航框架主要包括三大技术:地图构建,机器人定位和导航规划。这些技术在家庭场景中均已经得到长足的发展和进步,但是将这些技术应用到大规模复杂动态的公共服务场景时,却面临着一些工程方法和理论上的困难。针对这些问题,本文提出了一些提出了新的方法或技术改进,具体的内容和创新点如下:1.机器人导航需要有详细的环境模型信息,因此适用于大规模环境的地图构建算法是机器人导航技术的基础工作。通常公共服务场景的面积远大于普通家庭环境,可以达到上万平方米甚至更大的规模。现有的地图构建算法在大规模场景地图构建问题中效率低下,需要耗费大量的存储资源。针对大规模的场景地图构建问题,本文提出一种基于四叉树的高效地图表示方法;并设计了一套应用于四叉树地图的访问码机制,利用该机制能够快速的访问地图节点数据;最后提出了大规模环境下基于四叉树地图表示的地图构建方法,实验表明基于四叉树的SLAM方法能够减少50%到70%的内存消耗。2.公共服务场景的环境十分复杂,具有高度的动态性,环境中通常存在密集且流动的人群。人群密集环境下的机器人导航依赖于稳定的机器人定位模块,但是复杂动态环境下的机器人定位十分困难。本文分析了动态环境下的机器人定位存在的问题及其原因,并设计了基于里程计标定和环境自感知的定位改进算法,极大地降低了错误观察对机器人定位的影响,从而显著提高动态环境下机器人定位的鲁棒性。3.传统的机器人导航系统缺乏对环境中行人因素的考虑,其导航表现并不能满足公共服务领域场景下Human-aware导航的需求,与人类自身的导航行为相差甚远。针对当前导航规划算法在人机环境下存在的问题,本文利用激光传感器对行人进行识别和跟踪,并对其运动状态进行估计和预测;进而利用行人预测信息来改进基于动态窗口的局部路径规划算法,提高机器人导航在人机交互方面的舒适性,自然性和交互性。总而言之,本文针对传统机器人导航技术在公共服务领域应用中存在的不足和问题,分别在地图构建、机器人定位和导航规划三个方面提出了新的算法或技术改进,并实现了一套适用于大规模复杂环境下的机器人导航解决方案,并在可佳商场导购机器人项目中得以实施。
[Abstract]:Robot navigation is a basic research field, its significance lies in: in the autonomous mobile environment is a prerequisite for many kinds of robots can perform other complex tasks. In recent decades, with the development of robot technology and artificial intelligence technology, the growing and the whole society of service robot needs academia and industry have invested a lot of resources for robot navigation technology is studied and applied to explore, make indoor robot navigation technology is becoming more and more mature. Indoor robot navigation technology can be divided into three stages: the stage of industrial robots, service robots in the home stage and public service robot scene public service stage. Compared with the factory scene and the ordinary family scene is more complex, this to the stage to adapt the ordinary family oriented indoor environment Robot navigation technology brings new problems and challenges, the background of this paper. The indoor robot navigation framework mainly includes three technologies: map building, robot localization and navigation planning. These technologies in the domestic scene had been considerable development and progress, but the application of these technologies to large-scale complex dynamic the public service scene, there are still some engineering methods and theoretical difficulties. To solve these problems, this paper puts forward some improvement put forward new methods or techniques, specific content and innovation are as follows: 1. robot navigation environment model need detailed information, so it is suitable for large-scale environment map building algorithm the basic work of the robot navigation technology. Usually public service scene area is much larger than the ordinary family environment, can reach tens of thousands of square meters or more The size of the existing map building construction algorithm. The problem of low efficiency in large scale scene map, requires a lot of storage resources. For the large scale scene map building problem, this paper proposes a representation method of efficient map based on the four fork tree; and design a set for the four fork tree map access code mechanism, the use of the mechanism can access the map node data quickly; finally proposes four tree map representation based on a large-scale environment map, experimental results show that the SLAM method based on the four fork tree can be reduced by 50% to 70% of the memory consumption of.2. public service environment is very complex, highly dynamic, intensive and the flow of the crowd usually in the environment. The crowded environment depends on the robot navigation robot positioning module is stable, but the robot in complex dynamic environment ten trapped Difficult. This paper analyzes the existing robot positioning under dynamic environment problems and their causes, and design improved self sensing positioning algorithm and based on the mileage meter calibration, which greatly reduces the error effect of robot localization, which significantly improve the robustness of traditional robot navigation system.3. robot positioning under dynamic environment the lack of pedestrian environment factors into consideration, the navigation performance can not meet the requirements of Human-aware navigation scene of public services, and a far cry from the navigation behavior of human beings. In view of the current navigation planning method in man-machine environment, by using the laser sensor recognition and tracking of pedestrians, and the movement of state estimation and prediction; and then use the pedestrian to improve the local path planning algorithm based on dynamic window information forecasting, improve robot navigation in people Machine interaction comfort, and interactive nature. In short, aiming at the shortcomings and problems of traditional robot navigation technology applied in the field of public service, respectively in map construction, robot localization and navigation planning three aspects put forward the improved algorithm or new technology, robot navigation and implements a set of suitable for mass under the complex environment solution, and can be implemented in the good shopping guide robot project.

【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP242

【相似文献】