移动式家居服务机器人的自主导航研究及实现

发布时间：2018-05-21 17:04

本文选题：同步定位与地图构建 + 机器人运动模型　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着机器人领域的相关技术持续发展,机器人技术中,同步定位与地图构建(SLAM,Simultaneous Localization and Mapping)获得了研究者的高度重视。快速并且鲁棒的特征提取、精确并且稳定的特征匹配、有效的闭环检测以及高效的全局优化算法,这些技术一直是这个领域研究探索的重中之重;然而机器人技术在工程方面的应用主要集中在低廉传感器情况下高精度并且高鲁棒性的设计与应用。最近,Rao-Blackwellised Particle Filter(RBPF)已经变成了一种解决SLAM问题很有效的方法。此方法使用粒子滤波,其中每个粒子承载了一个地独立的环境地图。因此,关键问题是如何减少粒子的数量。对于局部路径规划,执行避障时,传统的Elastic Band算法依据最短路径的思想改变由全局规划器生成的路径。但是此方法仅仅考虑了环境的动态性,而没有考虑任何机器人相关的动态约束,因此产生的局部轨迹不是最优的。本文基于基本的Rao-Blackwellised Particle Filter(RBPF)算法给出了一个计算精确的提议分布的方法,此方法考虑了机器人的运动状态和最近时刻的观测值。它有效减小了预测过程中机器人位姿的不确定性。此外,设计了一个自适应重采样操作,这很大程度上减小了粒子退化的风险。针对EB算法的问题,本文中给出了“Timed Elastic Band”(TEB)方法。此方法通过改变由全局规划生成的初始路径优化机器人局部轨迹。在轨迹优化中所考虑的目标函数包括轨迹执行时间,与障碍物之间的距离,通过的路径点和需要遵守的机器人动力,运动和几何约束。TEB明确地考虑运动的时间和空间方面,例如受限的机器人速度和加速度。轨迹规划算法实时的执行,例如TEB实时的处理动态的障碍物和运动约束。TEB问题公式化为一个量化的多目标优化问题。大多数目标函数是局部的并且仅仅和几个连续的机器人状态有关。这个局部性产生一个稀疏的系统矩阵,因此可以使用快速的并且有效的优化方法例如G2O框架解决TEB问题。G2O稀疏系统解算器已被应用到视觉SLAM问题中。这有助于G2O框架在TEB轨迹改变中的应用。最后在实际环境中验证说明文中使用的算法是鲁棒并且很有效。为了实现机器人的完全自主性,本文中研究了一种基于边缘概念的自主搜索方法,边缘位于开放空间和未搜索空间之间的边界上。通过移动机器人不断地移动到新的边缘,机器人能够持续地扩展地图,直到环境中所有区域均被搜索为止。最后,通过仿真和使用真实的机器人进行实验获得的结果可知,均说明了改进的RBPF SLAM算法在减少粒子数方面的优势,表明了TEB算法是鲁棒的并且能够实时的生成有效的局部最优轨迹,证明了基于边缘检测搜索算法相对于沿墙行走的优势。
[Abstract]:With the continuous development of the related technologies in the field of robotics, synchronous localization and map building have attracted great attention from researchers. Fast and robust feature extraction, accurate and stable feature matching, efficient closed-loop detection and efficient global optimization are the most important technologies in this field. However, the applications of robotics in engineering are mainly focused on the design and application of high precision and robustness in the case of low cost sensors. Recently, Rao-Blackwellised Particle filter has become an effective solution to the SLAM problem. This method uses particle filtering, where each particle carries an independent map of the environment. The key question, therefore, is how to reduce the number of particles. For local path planning, when obstacle avoidance is performed, the traditional Elastic Band algorithm changes the path generated by the global planner according to the idea of the shortest path. But this method only considers the dynamic nature of the environment and does not take into account any dynamic constraints related to the robot, so the local trajectory generated is not optimal. Based on the basic Rao-Blackwellised Particle filter (RBPF) algorithm, this paper presents a method to calculate the exact proposed distribution, which takes into account the motion state of the robot and the observations at the nearest moment. It effectively reduces the uncertainty of robot pose in the prediction process. In addition, an adaptive resampling operation is designed, which greatly reduces the risk of particle degradation. In order to solve the problem of EB algorithm, the "Timed Elastic Band" method is presented in this paper. This method optimizes the local trajectory of the robot by changing the initial path generated by the global planning. Objective functions considered in trajectory optimization include trajectory execution time, distance from obstacles, passing path points and robot dynamics to be observed, motion and geometric constraints. TEB explicitly considers the temporal and spatial aspects of motion. For example, the speed and acceleration of a restricted robot. The real-time execution of trajectory planning algorithms, such as TEB real-time processing of dynamic obstacles and motion constraints. TEB problem formulated as a quantized multi-objective optimization problem. Most objective functions are local and relate only to several consecutive robot states. This locality produces a sparse system matrix, so fast and efficient optimization methods such as the G2O framework can be used to solve the TEB problem. The G2O sparse system solver has been applied to the visual SLAM problem. This is helpful to the application of G 2O framework in TEB trajectory change. Finally, the algorithm used in the actual environment is proved to be robust and effective. In order to realize the complete autonomy of the robot, an autonomous search method based on the edge concept is studied in this paper. The edge is located on the boundary between the open space and the unsearched space. By moving continuously to the new edge, the robot can continuously expand the map until all areas in the environment are searched. Finally, the results of simulation and experiments with real robots show that the improved RBPF SLAM algorithm has the advantages of reducing the number of particles. It is shown that the TEB algorithm is robust and can generate effective local optimal trajectory in real time. It is proved that the edge detection based search algorithm is superior to the moving along the wall.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【相似文献】