【作者】 尚文；

【导师】 戴先中； 马旭东；

【作者基本信息】 东南大学 ， 控制理论与控制工程， 2005， 博士

【摘要】 网络机器人系统强大功能的实现离不开机器人本体的智能行为能力。本文以室内环境下网络化移动机器人开发应用为背景,对移动机器人在基于特征的室内环境中实现高性能的自主定位导航和环境建模等问题进行了深入研究。环境特征的描述与识别是基于特征的室内办公室环境中机器人定位导航和环境建模的基础。环境特征主要从声纳和视觉等传感信息中提取,针对声纳信息单个声纳数据抗干扰性差的缺点,提出了改善传感性能的声纳直线特征提取方法;考虑到室内环境中多面体占主导的情况,提出融合颜色模型和感知组织的贝叶斯网络识别和描述三维多面体特征的方法,并给出了典型特征的识别结果;文中还对较为简单的二维平面特征(轨线)提取直线特征,并实现了简单而实用的轨线跟踪导航系统实例。对于已知环境模型下的机器人全局定位,要求保证鲁棒的定位性能和精确的位姿跟踪能力,这是实现室内环境中自主导航的关键。针对Markov方法和Kalman滤波器两种概率定位方法存在的不足,提出并实现了Markov-EKF概率定位方法,兼具了全局定位鲁棒性和位姿跟踪精确性。通过路径跟踪实验以及与Markov方法的定位误差比较,验证了定位方法的性能。为了增强在不确定性条件下定位导航的抗干扰性能,同时解决机器人被绑架问题,以Monte Carlo方法为基础并进行改进,提出了基于视觉特征的扩展Monte Carlo方法。实验验证了对于未建模事件(如初始定位、绑架诱拐等)的处理能力。另一方面,未知模型环境下在线建模—SLAM(同时定位与环境建模)是移动机器人实现自主能力的又一重要问题,其关键在于计算复杂性和适用的环境。文中针对这一问题提出了集成几何方法和拓扑方法的分级混合建模方法,用来构造大规模的环境,生成全局拓扑结构和局部基于特征模型的混合等级式模型,减小了模型构造的计算复杂性,提高了整个环境模型的准确度。小规模仿真环形环境、小规模实际环境和大规模环境中的建模结果验证了分级混合建模方法的有效性。通过网络环境下基于视觉的轨线跟踪、全局定位和拓扑建模应用,将以上智能行为应用到网络机器人系统中,验证了智能行为与网络机器人系统集成应用的可行性和有效性。更多还原

【Abstract】 Powerful applications of the networked robotic system greatly depend on intelligent behavioral abilities of the mobile robot itself. For more powerful applications of networked robotic system, researches on the autonomous localization, navigation, and environmental modeling of mobile robot in feature-based indoor environments are presented in this dissertation.Description and recognition of features, normally extracted from sonar or vision sensors, are fundamental for robot navigation. To avoid the poor characteristics of a single sonar data, sonar line features are extracted to improve performance. And as polyhedron features prevail in office indoor environments, Bayesian networks recognition method is introduced which combines color models and perceptual organization, and recognition results of typical features indoor are given. Additionally, 2D planar line features are extracted with a simple and practicable tracking navigation system realized.On one hand, global localization is the most important behaviors for autonomous navigation of robot, for which robustness and precision a necessity. To enhance the performance of Markov and Kalman filter for application, Markov-EKF is proposed for robust global localization and precise pose tracking. Path tracking experiment validates the performance of Markov-EKF method with localization error comparison with Markov method given. In order to improve the localization performance in uncertain or un-modeled conditions, as well as to resolve the kidnapped problem, vision-based extended Monte Carlo algorithm is proposed. Localization experiments demonstrate the ability to deal with the un-modeled events.On the other hand, Environmental modeling on-line—SLAM(Simultaneous Localization and Mapping) is another important problem for mobile robot. A hybrid hierarchical mapping method is proposed which integrates topological and feature-based method to extend to large scale and cyclic environments, thus reducing the computation complexity and improving the precision of environment model. Experimental results in typical environments, small simulating cyclic environment, small real environment and large scale environment, all demonstrate the validity of the hierarchical method.Example applications of the intelligent behaviors above to networked robotic system, such as trajectory tracking, global localization, and topological mapping, validate the feasibility and effectiveness of integration and approaches.更多还原