A Robot Navigation System in Complex Terrain Based on Statistical Features of Point Clouds

The robot navigation system is mature in 2D flat terrain, however, in 3D complex terrain scenarios the computational burden increases dramatically because an additional dimension is added to the mapping and path planning problems. To this end, a hierarchical controlled robot navigation system in complex terrain (HCRNS-CT) is proposed for mapping and path planning by using a 2D grid map based on statistical features of the height of point clouds to replace the ordinary 3D map and implemented in a mobile platform in this article. In HCRNS-CT, a 2D grid map that can reflect the curve of the ground is proposed from the viewpoint of statistical features. These point cloud features from the depth camera can be used to represent the traversability of each grid, even if the grid size is large, and thus can reduce the time consumption on mapping and path planning dramatically. And Kalman filter is used to update the global map with these features by local point clouds perceived by the depth camera. In addition, to match the introduced map, we propose a path planning algorithm fully utilizing those features and considering the mechanical structure of the robot. In HCRNS-CT, a hierarchical control structure is adopted as its backbone to reduce the frequency existing in the high computational burden module, and thus maintains the real-time standard of control. Therefore, HCRNS-CT can handle slopes and obstacles effectively in complex environments. A series of real-world experiments validate the robustness and effectiveness of HCRNS-CT.