激光导航AGV在特征地图中的全局定位方法研究

发布时间：2018-12-10 19:00

【摘要】：自动导引车(Automated guided vehicle,AGV)在智能制造与物流系统中得到了快速的发展,其全局定位是自主导航技术中的研究热点之一。马尔可夫(Markov)定位算法是一种基于概率分布的全局定位方法,其通用性强并能解决多模和非线性问题,得到了广泛的关注和研究。目前在特征地图中还没有适用的基于概率的全局定位方法,本文在建立AGV的Markov定位方法相关模型的基础上,对特征地图中的全局定位问题展开了研究。针对特征地图中应用Markov定位算法在对自动导引车全局定位时,常会出现传感器观测与地图之间的特征数据关联不唯一而导致定位失败的问题。提出了一种不通过数据关联的Markov定位计算新方法。利用高斯核函数将环境中的稀疏特征拟合成平滑致密曲线,通过对比传感器观测和算法预测得到的两个致密曲线相似度来计算Markov定位中的观测模型。同时直接利用电子罗盘传感器得到AGV的姿态信息,使算法只关注于求解AGV离散化位置的信度而不必同时计算AGV位姿的三维数据,在解决常规Markov定位方法在中心对称环境中失效问题的基础上减少了算法计算量,通过仿真分析验证了该全局定位方法的有效性。针对Markov定位算法计算量大、效率低的问题,在AGV姿态由电子罗盘信息直接得出的基础上提出了基于四叉树模型的变分辨率离散化平面栅格方法。通过将AGV位姿估计的三维状态空间减少到二维平面栅格,并在定位过程中减少对地图中信度常为零的栅格区域的重复计算以提高算法的运算效率和信度极值收敛速度。仿真结果验证了变分辨率栅格离散化方法的有效性,与基于高斯核函数的固定分辨率Markov定位方法相比,该方法对AGV的全局定位效率更高。最后,通过半封闭环境下的自动导引车全局定位实验,对比扩展卡尔曼滤波方法估计出的AGV运动轨迹,验证了本文方法即使在AGV初始位姿未知的情况下,估计出的AGV运动轨迹精度依旧更高,定位结果更有效。
[Abstract]:Automatic guided vehicle (Automated guided vehicle,AGV) has been developed rapidly in intelligent manufacturing and logistics system, and its global positioning is one of the research hotspots in autonomous navigation technology. Markov (Markov) localization algorithm is a global localization method based on probability distribution. It has strong generality and can solve multi-mode and nonlinear problems. At present, there is no probabilistic global localization method in feature map. Based on the establishment of AGV's Markov localization model, the global localization problem in feature map is studied in this paper. When the Markov algorithm is applied to the global localization of the automatic guided vehicle in the feature map, there is often a problem that the correlation between the sensor observation and the map is not unique, which leads to the failure of the location. A new method of Markov location calculation without data association is proposed. By using Gao Si kernel function, the sparse features in the environment are combined to form smooth dense curves, and the observation models in Markov location are calculated by comparing the similarity of the two dense curves obtained by sensor observation and algorithm prediction. At the same time, the attitude information of AGV is obtained by using the electronic compass sensor directly, so that the algorithm only focuses on solving the reliability of the discrete position of the AGV without having to calculate the three-dimensional data of the AGV position at the same time. On the basis of solving the failure problem of conventional Markov localization method in centrosymmetric environment, the computational complexity of the algorithm is reduced, and the effectiveness of the global localization method is verified by simulation analysis. Aiming at the problem of large computation and low efficiency of Markov localization algorithm, a variable resolution discrete plane grid method based on quadtree model is proposed based on the AGV attitude obtained directly from electronic compass information. By reducing the three-dimensional state space of AGV position and pose estimation to two-dimensional plane grid and reducing the repeated calculation of grid region where the information degree of map is usually zero in the process of location, the computational efficiency and the convergence speed of reliability extremum of the algorithm are improved. The simulation results verify the effectiveness of the variable resolution grid discretization method. Compared with the fixed resolution Markov location method based on Gao Si kernel function, this method is more efficient for global positioning of AGV. Finally, by comparing the AGV trajectory estimated by extended Kalman filter in semi-closed environment, it is verified that this method is even if the initial position and orientation of AGV are unknown. The estimated AGV trajectory accuracy is still higher and the localization result is more effective.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】