海底地形匹配高效质点滤波导航方法

Efficient Point-mass Filter Method for Seabed Terrain Aided Navigation

摘要:

海底地形匹配导航技术可实现自主水下机器人(Autonomous Underwater Vehicle，AUV)长时间的水下精确导航。质点滤波器(Point-Mass Filter，PMF)是实现 AUV 精确海底地形匹配导航的重要方法，但质点权重计算过程中需要寻找空间分布不规则的实测地形数据点与网格化先验地形图数据点间空间对应关系，造成大量计算消耗，严重影响了地形匹配导航算法实时性能。提出一种海底地形匹配高效质点滤波导航方法，通过构建伪输入高斯过程(Sparse Pseudo-input Gaussian Processes，SPGPs)模型，实现实测海底地形数据高效、精确网格化插值过程，并根据 SPGPs 方法提供的插值置信度实现网格化概率地图构建；搭建质点滤波器算法框架，提出适用于概率地图的质点权重计算过程，通过考虑网格化地图不同节点插值置信程度差别，提高地形匹配导航精度。于青岛中沙礁展开海上数据获取试验，并利用海试数据完成算法回放式仿真验证。试验结果证明，所提出网格化插值方法可实现高精度、高效地形深度插值过程，所提出海底地形匹配高效质点滤波导航方法可为 AUV 在线提供精确导航结果，且实时导航圆概率误差小于 3 m。

关键词:海底地形匹配导航;水下机器人;质点滤波;伪输入高斯过程

Abstract:

Terrain aided navigation can yield accurate navigation results for the long-endurance operation of autonomous underwater vehicles(AUVs). Point-mass filtering(PMF)theory is an important method to yield accurate terrain aided navigation results for AUVs，but the search for the spatial relationship calculation between the irregularly distributed bathymetric measurements and the gridded prior topographic map in mass point weight calculation will lead to a substantial amount of computational cost and severely weaken the real-time performance of the terrain matching navigation algorithm. An efficient PMF method for seabed terrain aided navigation is proposed in this paper. The proposed method yields efficient and accurate grid interpolation results for bathymetric measurements by constructing an Sparse Pseudo-input Gaussian Processes(SPGPs)model and builds a gridded probabilistic map using the beliefs of these interpolation results. The framework of a PMF is constructed，and a mass point weight calculation method using a probabilistic map is proposed to improve the positioning accuracy of the terrain aided navigation system by considering the difference in the interpolation confidence of nodes in the grid map. An at-sea data collection experiment was conducted in Zhongsha Reef，Qingdao，and play-back experiments were conducted by loading these field data. The experimental results showed that the proposed method can provide accurate interpolation results efficiently，and the proposed terrain aided navigation method can yield accuratenavigation results for AUVs with a real-time location error less than 3m(CEP).

Key words:terrain aided navigation;underwater vehicle;point-mass filter;sparse pseudo-input Gaussian processes