【作者】 吴月华；

【导师】 孟庆浩；

【作者基本信息】 天津大学 ， 检测技术与自动化装置， 2007， 硕士

【摘要】 移动机器人是一种在复杂的环境下工作的具有自规划、自组织、自适应能力的机器人。移动机器人在未知环境下的导航问题是制约当前移动机器人大规模应用的瓶颈。未知环境导航的其中两个关键问题是环境地图的实时建立和机器人的自定位。本文将重点放在了室内环境下机器人自身位姿和环境模型均未知情况下的移动机器人导航问题,即所谓的并发环境建模与定位问题(SLAM:Simultaneous Localization And Mapping)。论文首先介绍了SLAM问题涉及的两个关键技术,即环境地图的典型表达及机器人位姿的不同估计方法。其次结合超声测距特性,给出了基于旋转超声的贝叶斯网格地图的构建过程,并通过霍夫(Hough)变换改进了地图构建方式,增加了建模精度。本课题选用粒子滤波作为定位方法,文章介绍了粒子滤波定位的算法,并将直方图匹配引入到定位算法的传感器更新阶段,提高了粒子滤波算法的效率和鲁棒性。论文从降低数据关联的复杂度、提高实时性的角度出发,在贝叶斯栅格地图和粒子滤波定位这两种技术的基础上,提出了改进的基于粒子滤波的SLAM算法。改进的方法降低了位姿精度和地图精度的关联程度。此法在利用粒子滤波定位时,只需要在机器人附近分布粒子就能满足算法需要,减少了粒子数目,增强了实时性,仿真和实验结果均表明它是一种可行的方法。更多还原

【Abstract】 Mobile robot is a kind of robot which works in complicated environments and has an ability of self-programming, self-organizing and self-adapting. Mobile robot navigation in unknown environment is the bottleneck of using mobile robot broadly. The two key questions of navigation are building map in real time and obtaining robot’s pose precisely. This paper mainly focuses on mobile robot navigation in indoor environment where neither the pose of the robot nor the environment information is known, i.e. the so called Simultaneous Localization and Mapping (SLAM) problem.Firstly this paper introduces the two key techniques of SLAM, one of which is representative expression means of environmental map, and another is typical localization methods of mobile robots. On the basis of introducing the characteristic of sonar, the process of building grid map using Bayesian rule is presented. In order to increase the precision of the built map, Hough transform is applied. Particle filter is used to localize robot and its algorithm is explained. A histogram matching is utilized in the sensor-data-update phase of particle filter localization. Experiments show it makes the algorithm more efficient and robust.To reduce the complexity of data association and enhance the real-time performance of the algorithm, an improved SLAM algorithm is put forward using Bayesian gird map and partial filter. This algorithm reduces the association between the precision of map and that of robot pose. Particles are distributed only around the robot, so it promotes the real-time performance. Simulation and experiments results validate the feasibility of algorithm.更多还原