【作者】 潘薇；

【导师】 蔡自兴；

【作者基本信息】 中南大学 ， 控制科学与工程， 2009， 博士

【摘要】 随着机器人技术在各个领域的广泛应用和发展,在研究和应用双重需求的推动下,多机器人系统的研究已受到国内外研究结构和产业界的重视,并逐步成为一个充满活力、充满挑战的领域。在一些面向任务的应用中,地图构建是实现多机器人系统自主导航、在未知环境中完成复杂智能任务的关键,也集中体现了多机器人系统的感知能力和智力水平。因此,如何利用多个机器人来构建精确有效的环境地图,无疑是多机器人研究中一个重要且关键的课题。本文围绕多机器人研究中基本而关键的建图问题,深入研究了多机器人群连续避障和避碰策略、基于相对观测量的同时定位与建图(SLAM)方法以及局部地图融合等相关内容。论文首先研究了多机器人建图过程中的避障和避碰问题。针对声纳机器人建立了基于避障子行为状态的方向选择规则,配合检测到的当前障碍物方位,来选择正确的避障方向;在多机器人避碰策略上,建立了一种无交通灯的交叉路口避碰模型,在此模型中,设计了优先通过权评价函数以决定机器人谁将获得优先通过交叉路口的权利。多机器人系统的整体避碰关系则利用有向图来描述,并通过逐步消除图中各个环路来解决死锁问题。其后提出了一种基于扩展卡尔曼滤波器(EKF)的多机器人同时定位与建图的新方法。该方法首先利用支持向量机对EKF-SLAM方法予以改进,根据新息相关性,自适应地调节测量噪声方差,有效地解决了噪声的统计特性与实际不符合时滤波器发散的现象,提高了SLAM精度。之后,将该种应用于单机器人的改进EKF-SLAM方法应用于多个机器人,在机器人会合时测量彼此之间的相对距离和角度,以利用该相对观测量进行局部地图的坐标转换与融合。考虑到传感器观测信息的不确定性会降低坐标转换精度,引入联合兼容分枝定界(Joint Compatibility Branch and Bound,JCBB)算法进行路标的增强匹配,利用匹配路标信息进一步降低地图融合过程中的误差和计算复杂度。与常规的多机器人SLAM方法相比,该方法无需已知机器人的相对起始位置信息,不要求各机器人构建的局部地图重叠,限制条件少,在实际应用中具有更大的灵活性。针对多个声纳机器人,研究了基于粒子滤波器的多机器人FastSLAM方法。首先将常规粒子滤波器与粒子群优化算法有机结合,引入最新的机器人观测信息以调整粒子的提议分布,从而在保证算法精度的同时,极大地减少了定位与建图所需的粒子数,并有效缓解了粒子退化现象。此外,考虑到常规的重采样过程容易引起样本贫化现象,引入了概率算子以增加粒子的多样性。之后,将该种改进的FastSLAM方法扩展到多机器人系统中,利用机器人会合时的相对观测量来初始化粒子滤波器,并利用虚拟机器人将所有机器人在会合前后的观测值融合成为全局栅格地图。实验结果表明该方法具有较高的精度、稳定性以及灵活的地图表示方式。考虑到在实际环境中,机器人可能无法满足至少相遇一次的条件,提出一种基于栅格融合的多机器人建图方法。该方法让各机器人独立探索环境并对不同的局部栅格地图予以融合。在地图融合过程中,无需考虑机器人相对位置的先验信息,而是以栅格地图相似度为度量标准,利用距离变换和改进的遗传算法高效、快速地搜索各局部地图之间的最大重叠部分,进而予以融合。实验结果表明,该方法无需考虑机器人的位置信息,限制条件少,更适合于声纳机器人的实际建图应用。在论文的最后,总结了整篇论文的工作,指出了进一步研究探索的方向。更多还原

【Abstract】 Robotics has already been widely developed and used in many fields. Urged by investigation and application, research of multi-robot system has received more and more attention, and it is gradually becoming an active field full of challenge. In some tasks, such as military operation, aviation, services and RoboCup, map building is not only very important to accomplish autonomous navigation and other complex intelligent tasks, but also embodies perception ability and intelligence of multi-robot systems. So it is an important and key problem of research on multi-robot to building environment map accurately and effectively by multiple robots.This work focuses on the basic and important problem in multiple robots research—multi-robot map building. It addresses the problem of continuous obstacle avoidance and collision avoidance strategy, simultaneous localization and mapping (SLAM) based on relative observations and local map merging of a mobile robot team.First, the problem of multi-robot obstacle avoidance and collision avoidance is discussed. Aiming at sonar robots, a kind of direction selection rules based on the avoiding sub-behavior states is designed, to choose the right avoiding direction, with the cooperation of the current obstacle orientation been detected. For the collision avoidance strategy of multi robots, a model named "No-Traffic-Light-Crossing" is built. In this model, a Pass-Priority evaluating function is designed, to decide which of the robots in the model will gain the pass priority. Digraph is used to describe the avoidance relation of multi-robot system, and the "dead-lock" problem will be solved through stepwise eliminate the cycles of the digraph.Second, a novel approach to multi-robot simultanieous localization and mapping (SLAM) based on Extended Kalman Filter (EKF) is presented. The EKF-SLAM approach is improved by support vector machines. According to relativity of innovation, the measurement noise covariance was adjusted adaptively. The method could cope with divergence problem caused by the insufficiently knowing of the prior filter statistics and improve the accuracy of SLAM. Then, the improved EKF-SLAM approach used in a single robot in common was extended to a multi-robot system. Relative observations between robots when meeting was processed to execute coordinate transformation and map merging. In addition, by considering the uncertainty of sensor information, an improved matching of landmarks was introduced to the process of map fusion to increase accuracy of localization and mapping. Compared with typical ones, this approach can accomplish multi-robot simultaneous localization and mapping accurately and effectively without using any initial pose information of robots and no need for robots to explore duplicate areas, so as to be more suitable for a variety of complex cases in application with less restrictions.Aiming at multiple sonar robots, a multi-robot FastSLAM approach is discussed. First, the standard particle filter and particle swarm optimization algorithm are incorporated into the filtering framework of this approach. The newest observations are introduced to adjust particles’ proposal distribution, so as to largely reduce the sample size necessary for localization and mapping and effectively relieve the particle degeneracy problem while ensuring the algorithm precision. In addition, considering that the typical resampling process always leads to the loss of diversity in particles, a probabilistic operator is introduced to keep the diversity of particle swarm. Second, the improved FastSLAM approach is extended to a multi-robot system. The relative observations when robots’ meeting is used to initialize the particle filter, and the subsequent and prior observations from robots are combined into a global grid map by using virtual robots. Experimental results show the approach has high accuracy and stability as well as flexible map representations.Considering that the robots may never encounter at all in real environment, a multi-robot mapping approach based on grid matching is presented. The approach lets all robots operate individually and then tries to merge the different local grid maps into a single global one. Without using any pose information of robots, the process of map merging is performed by measuring the similarity between grid maps. Distance transforms and an improved genetic algorithm are used to effectively search the maximum overlap at which the local maps can be joined together. Experimental results show that this approach can accomplish multi-robot map building accurately and effectively without using any robots’ position information, and be more suitable for sonar robots in applications for mapping with less restrictions.Finally, we summarize the general work of this thesis and give a short outlook on possible future research.更多还原