3D Map Construction In Unknown Indoor Environment Using Mobile Robot Equipped With3D-LRF

Mobile robot is a comprehensive system of multi-field cross withmultidisciplinary theory, which combines artificial intelligence,environmental detection, path planning, control and execution moduledesign, mechanical design, and many other functional requirements into anintegrated system. It also needs to synthesize latest research on engineerinformation processing, electronic circuit design, computer science,intelligent control, artificial intelligence and many other fields, whichreflects the highest level of national integrated mechatronic design andmanufacturing capabilities. To make the mobile robot explore in a complexenvironment automatically, the environmental information around theobtained through sensors, then conduct explicit perception and analysis onthese sensor data. Mobile robot depends on map construction to a very largeextent, thus study the approaches to created large-scale precise3D(three-dimensional) map has extremely vital significance.Traditional map creation using mobile robot is usually realized throughmonocular or binocular vision camera or2D laser range finder, but cameracould not work when ambient light changes or the too complex environment,two-dimensional laser range finder can only detect environmentalinformation at a certain height, thus these robots can only move aroundsimple environments, because two-dimensional map cannot effectivelydescribe the3D structure in space. In recent years, Kinect sensor provides afast and efficient way to obtain3D point cloud data, but considering itsprecision, it is still not suitable for the creation of large-scale precise3D map.In order to achieve the goal of building3D map in an unknown complex indoor environment, we adopted a3D laser range finder composed by a2Dlaser range finder and a rotating platform, detailed study and experimentsare carried out for a series of questions during map construction process.Firstly, we use3D laser range finder to collect360°point cloud data,then filter the data through statistical methods to trim the noise and floatingpoint, this step is followed by a down-sampling processing which removesuseless points to make the raw data more effective. Likelihood functionbased registration is then conducted for the3D point cloud stitching whichproduces a complete3D point cloud map. Because the storage capacity for3D point cloud data is too large, we use the Octree data structure to convert3D point cloud map into a3D grid map. We improved D*algorithm basedon this data structure and realized path planning under3D grid map. In orderto let robot determine the next target location independently, we reduce the3D map into two-dimensional in order to improve the algorithmperformance, then use frontier based method to search for border area,according to the known and unknown region information. We also carriedout the active closed-loop detection to improve the accuracy of map andpositioning, robot monitors the shortest distance between current locationand the previously accessed location in order to perform closed-loopoperation. Experiments and practical tests using real mobile robot equippedwith these methods have been carried out in Shanghai Jiaotong UniversitySiyuan Building, the detailed data and experimental conditions has beenrecorded and displayed in colored pictures, the time and accuracy of3D mapshowed that this map construction method is effective and feasible, which issignificant to the promotion of mobile autonomous mobile robots.