Research On Three-dimensional Mapping Methods For Autonomous Navigation Of Robots

Mobile robot is an important subclass of robotics,and has been applied in various fields,such as military and service industries.Since they should navigate in unpredictable,dynamic and unstructured working environments to perform most of tasks,autonomous navigation is a fundamental and significant technology for mobile robots.Research on three-dimensional(3D)mapping methods for autonomous navigation of robots is therefore of theoretical and practice significance.One important approach to solve autonomous navigation is Simultaneous Localization and Mapping(SLAM).However,maps built by existing SLAM systems generally cannot be directly applied in navigation of mobile robots.And they should be transformed into other forms that are suitable for navigation,such as grid maps.Besides,SLAM system should be updated iteratively to make maps they built more accurate.Therefore,a real-time 3D grid mapping method is needed,to handle the updates of map build by SLAM system and transform it into grid map in real-time.While there is a real-time 3D grid mapping method,the efficiency of mapping and navigation of mobile robots could be low due to the limited computation and storage ability of mobile robots and large-scale environments.As for ground mobile robots,they do not have full 3D manuverbility like aerial robots or underwater robots,so that they cannot make full use of information provided by full 3D representations,which typically have high computational and storage demands.Therefore,an efficient and compact spatial representation of the environment for navigation of ground robots is required.Focusing on these problems,we first propose a real-time globally consistent 3D grid mapping method,to produce 3D maps which are suitable for navigation of robots in real-time.We also propose a real-time update strategy and an efficient data structure based on identity number of keyframe.Experimental results in static and dynamic scenarios demonstrate that the 3D mapping method can produce globally consistent 3D grid maps,which can be directly applied to path planning of mobile robots.An efficient and compact spatial representation has also been proposed in this paper,which is suitable for navigation of ground robots and is named 2.5D normal distributions transform map(2.5D-NDT map).The representation is following the central idea of normal distributions transform.It only stores space information that are potentially useful for navigation of ground robots.And a two-layer index structure has been proposed to efficiently manage the map.Moreover,algorithms for traversability analysis and path planning have been proposed.Experiments demonstrate that the proposed representation can represent 3D environments properly and compactly for navigation of ground robots.Moreover,paths suitable for navigation of ground robots can be planned efficiently based on proposed algorithms.Comparative experiments show that the decrease in storage space required by the 2.5D-NDT map reaches 63%-80%,while the reduction ratio in time consuming by generating the 2.5D-NDT map reaches 57%.To sum up,based on above two works,mobile robots can build globally consistent maps that are suitable for navigation.Besides,utilizing the 2.5D-NDT map,ground mobile robots can navigate autonomously in large-scale environments in real-time.