【作者】 陈杭；

【导师】 王伟；

【作者基本信息】 大连理工大学 ， 控制理论与控制工程， 2009， 硕士

【摘要】 移动机器人对环境的有效辨识是其自主导航和环境探索的关键,有效的感知环境信息并构建环境模型能够保证移动机器人安全合理的进行运动规划。与结构化场景相比,存在废墟、复杂地形以及动态障碍物等复杂环境信息的三维非结构化场景对环境的建模和重构以及在此基础上的运动规划提出了更高的要求。本文的研究重点为针对三维非结构化环境的建模与自主运动规划,旨在通过这两方面的研究建立一套具有针对三维非结构化环境实时建模和重构,并以此为基础进行实时规划和决策功能的移动机器人智能控制系统。对于由三维测距系统获取的激光点云数据,本文以多种方式实现了三维场景的建模,其中包括直线特征提取,平面特征提取及基于高程模型的三维地图构建等。在平面特征提取中,针对不同的应用,利用两种不同的方法有效的实现了平面分割。在区域扩张法平面分割算法中,利用法向量对生成平面进行校正,有效的克服了区域扩张中存在的平滑渐变问题。在构建出的环境模型的基础上,提出了两种自主运动规划的方法:基于几何-高程模型的自主运动规划方法和基于地形评估的自主运动规划方法。前者利用一种综合了高程模型和平面模型的几何-高程模型来表述非结构化场景并构建出三维随机路图,之后利用Dijkstra算法搜索一条满足安全性的最短路径。这种方法克服了高程模型和平面特征模型各自存在的环境信息不完整的缺点,增强了环境模型的有效性和路径规划的合理性。后者利用水平栅格将整个场景离散化,并利用栅格内和栅格间点云高度分布情况对地形进行评估,同时根据机器人的通过能力将整幅场景细分为多个可行区域。依据评估后的场景,提出了一种满足多种约束并可适应不同规划级别的自主运动规划方法,增强了机器人自主运动规划的适应性和鲁棒性。实验结果证明了这两种方法的有效性和实用性。更多还原

【Abstract】 The effective recognition and perception of 3D environment are the most fundamental problems that have to be solved before the mobile robot can navigate and explore autonomously in complex outdoor environment. Compared with the indoor environment, the outdoor environment with completely unstructured characteristics has put forward higher requirements for 3D environment modeling and reconstruction as well as motion planning.This study focuses on 3D environment modeling and autonomous motion planning of unstructured scene, which aims to build a mobile robot intelligent control system with real time modeling and reconstruction ability as well as real time motion planning and decision-making function.For the laser range data acquired from 3D ranging system, different methods are applied to implement the modeling of unstructured scene in this thesis, which include line extraction, plane extraction and 3D map building et al. Two different methods are proposed to implement plane segmentation. In the area-expending algorithm, we solve the smooth-gradient problem effectively according to the angle between normal vectors.Based on the environment model, two different autonomous motion planning methods are proposed: motion planning based on geometry-altitude model and motion planning based on terrain evaluation. In the former method, we synthesize geometry model and elevation model to build a 3D road map, then acquire final path to meet the security require using Dijkstra algorithm, this method overcome their respective disadvantages of the two models effectively and enable the validity and rationality of the final path. In the latter method, we distribute the laser range data into horizontal grids, and make terrain evaluation according to the distribution of the laser range data in and between grids. During motion planning, the whole scene is divided into different area with corresponding traversable value, and then a motion planning method is proposed which has the advantage of meeting multiple goals and being adaptive to different situation. The experiment results show the validity and robustness of these motion planning methods.更多还原