Path Planning For Ground Mobile Robots Based On 3D Point Cloud Map

Path planning is the basis for ground robots to complete various autonomous tasks efficiently.Existing path planning methods for ground robots usually require preprocessing such as voxel simplification or advanced representation of the environmental map obtained by the sensor,and the processed map will lose some environmental information.Therefore,the generated optimal path still needs to be greatly adjusted when used.The point cloud map is the complete original representation of the environment,which contains rich environmental information.Therefore,this thesis mainly studies the path planning method of ground robots based on point cloud map,including the extraction and evaluation method of traversable area based on geometric features and the path planning algorithm combined with traversability.This thesis directly uses the 3D point cloud map,fully considering the terrain and other factors in the environment,to provide a safer and more reliable path for ground robots to use directly.This thesis proposes a path planning system for ground robots based on point cloud maps.The system consists of a hardware driver module,a map optimization module and a navigation planning module.The hardware driver module is mainly used to drive the lidar,and to control the ground robots according to the speed offered by the navigation planning module.The map optimization module is mainly used to extract the traversable area of the ground robots from the original point cloud map,and terrain assessment is performed in the traversable area to form a traversable evaluation map.The navigation planning module mainly uses the traversable evaluation map and the movement characteristics of the robot to generate a planning path,and uses the path tracking algorithm to calculate the speed and send it to the hardware driver module.According to the actual application requirements of ground robots,in the map optimization module,this thesis focuses on the extraction and evaluation method of the traversable area.First,this thesis uses the Le GO-LOAM algorithm to generate the 3D point cloud map of the environment.Then,this thesis completes the extraction of the traversable area based on the geometric features such as the point cloud distribution in the point cloud map and the movement characteristics of the ground robots.And this thesis uses the terrain flatness and boundary risk factor to describe the traversability of each terrain point to the ground robot.Besides,in view of the dense distribution of candidate points in the traversable evaluation map,this thesis studies a point cloud downsampling method based on traversability,which can reduce the redundancy of the map while preserving the terrain features.In the navigation planning module,the optimized traversable map is used for path planning.Combined with the traversability of each terrain point,the search-based A star algorithm and the sample-based RRT algorithm are improved to generate a planning path that satisfies the constraints of path length and traversablity quickly.For the smoothness problem of the path generated by the path planning unit,the path is smoothed by introducing the look-ahead distance into the path tracking unit,and a speed smoothing factor is used to make the robot move more smoothly.Tested in Stevens dataset and real campus environment,the experimental results show that the path planning system can quickly plan a path that conforms to the robots’ motion laws and geometric constraints directly on the point cloud map,and the physical robot can reach the target point through the given path safely and correctly.