【作者】 朱江；

【导师】 王耀南；

【作者基本信息】 湖南大学 ， 控制科学与工程， 2011， 博士

【摘要】 近年来人们加快了使用移动机器人代替人类在危险、恶劣环境中作业的步伐,移动机器人的工作场合也越来越多地面向室外非结构化环境。野外、行星表面等室外非结构化环境具有多样性、随机性与复杂性等特点,相应地对移动机器人的自主性和智能化提出了更高的要求。而环境的认知和导航避障是实现移动机器人自主能力的基础性问题,同时也是关键性的难点问题。为此,本文主要研究了非结构化环境下的认知与导航避障方法,首先设计了可移植性高的自主导航硬件结构和软件系统,然后针对非结构化环境下的场景理解和自主导航方法开展了研究。全文主要工作包括如下几个方面。论文首先介绍了研究背景及意义,系统深入地概述了移动机器人的定义与发展、典型代表及其应用,在综述移动机器人导航技术的国内外研究现状的基础上,分析移动机器人自主导航中所面临的难点问题。在第二部分,详细介绍了本课题组自主研制的移动机器人自主导航系统及其软、硬件构成和基本功能。通过实际系统的开发,为本论文的算法研究提供了思路,并为算法的实际应用提供了实验平台。障碍物检测与识别是移动机器人理解环境的基础,障碍物检测与识别的准确性直接影响后续认知结果,而非结构化环境中存在的大量随机、多样因素对障碍物的检测与识别造成极大干扰,因此,论文第三章在分析了常用的非结构化场景分割方法的基础上,将非结构化环境中的障碍物检测转化为标记非目标区域的问题,提出一种结合二维图像、三维信息和相关向量机(RVM)实时检测与识别障碍物的方法。该算法首先将场景分为感兴趣区域和非感兴趣区域两个部分。然后,利用像素的三维信息对感兴趣区域聚类,得到独立的障碍物,采用空间关系和几何特征描述障碍物。针对障碍物识别对象数量巨大、形状千差万别、难以收集全样本等问题,通过训练相关向量机分类器实现对非结构化环境下的实时障碍物分类识别。在各种非结构化环境中的实验结果表明该方法鲁棒性高,能较好地适应场景中的复杂随机变化,准确、有效、快速地检测识别障碍物。不同类型的地表对机器人的移动性能的影响也不一样。通过视觉分析识别地表的类型是一种较常用的方法,其中图像特征提取是决定算法可靠性和实时性的关键和难点问题。在非结构化环境中光照、灰尘等随机性因素造成即使同一类型地表的外观也有较大差异,不同类型地表的外观有时候却很相似,进一步给基于视觉的地表类型识别方法带来了困难。论文第四章分析了视觉识别地表类型方法中常用的颜色、纹理特征,提出利用Gabor小波和混合进化算法优化选取地表图像特征,以最少的图像特征节点和Gabor小波尺度方向提取地表特征,在保证可靠性的同时,有效地减少了计算量。然后提出基于相关向量机神经网络的地表类型识别方法,具有较好的识别效果。当移动机器人在非结构化地形上运行时,机器人需要具备实时判断地形可通过难易程度的能力。与平整室内环境中的导航任务相比,界定非结构化环境地形表面上的不可通行区域更复杂,其可通过性取决于机器人的越障能力和地形的特点。论文第五章在模糊逻辑框架下,融合视觉和空间信息从粗糙度、开阔度、坡度、不连续度和地面硬度五个方面描述地形。首先定义了地形的粗糙度、开阔度、连续性和硬度。为了正确感知斜坡地形,分析了其描述模型。根据移动机器人从不同位置观测斜坡时深度信息的变化趋势,确定移动机器人观测斜坡的方向,提出了应用RBF神经网络估算地形坡度值的方法。在此基础上,运用模糊推理机推理得到地形可通行性评价,为后续的移动机器人导航提供了决策依据。为实现未知环境下对目标位置的无碰撞高效导航控制,论文第六章提出一种基于混合协调机制和分层结构的反应式行为导航方法。首先将行为分为高层行为和底层基本行为两层结构,提高控制器的实时性：然后设计了底层的模糊神经网络基本行为控制器,运用遗传算法优化调整控制器,以获得控制器更好的性能。行为的切换和调度分两层进行：根据神经网络辨识环境的结果选择高层行为；在底层通过计算当前环境与环境原型之间的匹配度确定融合基本行为的权值。在此基础上,融合地形可通行性分析结果,实现了非结构化环境下的自主导航避障。实验表明,该方法对不同环境有很好的适应能力和可靠性。论文最后总结了全文的主要工作和创新性的研究成果,并对下一步研究工作进行了展望。更多还原

【Abstract】 In recent decades, the pace which mobile robots are implemented their pace to expand human’s ability and release them from dangerous, hard and dirty work is quickening. More and more mobile robots work in outdoor unstructured environment. Because the unstructured environment, such as field environment, planetary surface et al, has the characteristics of variability, randomness and complexity, human expect more autonomous and intelligent mobile robot. The ability of environment cognitive and navigation avoidance are two basic elements in the autonomous mobile robot and key difficult points as well. This dissertation focuses on the methods of cognitive and navigation avoidance in unstructured environment. An autonomous navigation hardware and software with high portability are developed, then the challenging problems of scene understanding and autonomous navigation are addressed and some solutions are provided. Main results and contributions of this dissertation are as follows:Firstly, the background and significance of research work about this thesis are introduced. Then, the definition and development of mobile robot is surveyed, and some typical series of intelligent mobile robot and its applications are presented. And the research progresses of key technologies in mobile robot autonomous navigation in unknown environment are generalized. Furthermore, on the basis of the research, the difficult problems in autonomous navigation are discussed.In chapter 2, an autonomous navigation system is developed, and the software and hardware structure of this system is introduced. This system provides us ideas to present these methods in this dissertation, and it can be used as an experiment platform to verify methods.The performance of obstacle detection and recognition are two of basic problems for the mobile robot understanding environment, which affect the following sensing result directly. As the unstructured environment is usually ramdom and variability, it is difficult to detect and recognize accurately by the traditional method. In chapter 3, on the basis of analyzing traditional unstructured environment scene segementation methods, we propose an obstacle detection and recognition approach using image information,3-D information and relevance vector machine(RVM). Firstly, the scene is divided into two distinct regions:interest regions and uninterested regions. Then detected obstacle points are clustered into objects on the basis of their three-dimensional information, connectivity. The key obstacle features are described using geometric property. The RVM is used to categorizing the objects into three groups:stone, rock and slope. Extensive experiments show the utility and performance of the proposed approach. According to the large quantity and various shapes of obstacle, the real-time obstacle classification and recognition in the unstructured environment is realized by training RVM Classifier. In various unstructured environments, results of experiments show that this method is highly robustness, better adapt to random changes in complex scene.Different ground materials have different effect on the motion ability of mobile robots. To determine the surface types through visual analysis, feature extraction is the main difficulty which determine the real-time and reliability of the algorithm. In the unstructured environment, random factors, like illumination and dust, may resulting that the same type of surface appear very differently but different types of surface sometimes appear very similar, which brings further difficulties to the surface types identification based on vision method. In the fourth chapter, the paper analyzed the color, texture feature which commonly used in identification of surface type with vision method. An approach using Gabor wavelet and hybrid evolutionary algorithm optimize selection of surface image characteristics, extract surface features with fewest number of image features nodes and Gabor wavelet scales direction. Then, Relevance Vector Machines neruo-based approach to surface type identification is proposed. Results of experiments show that this method is reliability, and the computation is reduced as well.The mobile robot must have the ability of determining weather the terrain is traversable when operating on the unstructured terrain. Compared with navigation task in planar indoor environment, it is more complex to decide the untraversable region. The key terrain characteristics are identified as roughness, openness, slope, discontinuity and hardness. These characteristics are extracted from vision data and spatial information, and are represented in a fuzzy logic framework. Firstly, we define roughness, openness, discontinuity and hardness. In order to correctly perceive slope terrain, its discription model is discussed. Then, a novel method of estimating the slope of terrain using RBF neural network is proposed. Even on the condition that the position the robot observe the slope is unknown, the slope of the terrain can still be estimated correctly, the relative position between the robot and the slope can be obtained as well. Based on this method, we achieved the terrain traversability assessment through fuzzy inference machine, providing decision-making basis for subsequent mobile robot navigationIn order to realize the effective navigation of mobile robot to target position without collision, this thesis proposed a fast navigation and obstacle avoidance algorithm based on hybrid coordination mechanisms and hierarchical behaviors in chapter 6. This method has a hierarchical architecture that divides the controller into several smaller subsystems will reduce the negative effect that a large rule-based may have on real-time performace. Consequently, the neruo-fuzzy navigation behavior controllers are designed, and the performance of these controllers can be solved by using genetic algorithm. Then, the higher behavior is selected according to the result of identigied environment using neurel network. The weigh of fused basic behaviors are decided by computing the matching degree between the current environment and prototype environment. At last, fusing the terrain traversablity assessment, the autonomous navigation system is realized in the unstructured environment. Results of experiments have been shown that the introduced method has good navigation performance, and it is reliable to different environments.Finally, the main innovations of the dissertation are summarized, and the fields for further research are prospected at the end of the dissertation.更多还原