Research On Localization And Map Building For Robot Working In Orchard

Global Navigation Satellite System for agricultural robot localization and navigation has been successfully used for many years in broad-acre farmland.However,GNSS will fail to work in orchards since signal interference often occurs because of the line-of-sight to satellites get occluded by tree canopies.Thus orchard operating robot should seek reliable methods for localization and navigation which do not rely on GNSS.In the research areas of indoor mobile robots,map-based localization and navigation methods which are not dependent on the GNSS have been extensively studied and the technologies are mature.Therefore,this study uses our own developed agricultural mobile robot as the experimental platform and conducts orchard environment map building research for two types of maps which include geometric feature map and grid map.The ultimate goal of the future research is to realize map-based localization and navigation methods for orchard operating robot and get rid of dependence on the GNSS.The main research results and conclusions are as follows:(1)Highly unstructured orchard environment also includes structured cylindrical trunks.Therefore,this research extracts the center point of the trunk through laser radar data clustering and trunk circular arc feature detecting.The robot builds a map of center point features of trunks of a certain work space in the initial robot pose.This map is a kind of geometric feature map.(2)This study utilizes the map of center point features of trunks that built before and proposes a method for localization based on this map.In the study,water cedar scene and cherry and maple trees scene are used respectively to simulate orchard scenarios which fruit trunks are thick or thin.Experiments have successfully extracted the center point of the trunks in the predetermined work space and built maps of center point features of trunks.Then the study uses the maps to realize the inter-row localization of the robot and get access to the robot's position and orientation in real-time during inter-row exercise.This method well solves the inaccurate problem of odometer positioning due to odometer accumulated error when it is applied in orchard environment.(3)To solve the headland localization problem of orchard operating robot,this study proposes a method for headland localization based on the landmarks.The cherry and maple trees scene's experimental results show the effectiveness of this method.It hopes that headland localization,steering,navigation and other difficult headland issues of orchard operating robot can be inspired and learned from this method.(4)Stereo vision sensor can provide abundant information but has narrow field of view while laser radar has large detection range but can only get sparse data.According to the characteristics of two sensors,this research proposes a method to build integrate local grid map based on the local grid maps built respectively by the two sensors.Pear orchard scene's experimental result shows that this local grid map building method makes the two sensors complement each other.It can portray orchard environment's 2D details better and improve the reliability of map building.(5)This study carries out pose-based global grid map building research.By using robot's real-time inter-row pose which is obtained by the localization based on the map of center point features of trunks and the local grid map building result under different pose,the algorithm converts the local grid map into the global grid map in real time until the robot finishes exploring the predetermined work space.The cherry and maple trees scene's experimental results show this method can accurately describe the 2D distribution of trees in the predetermined work space.