Topological Map Construction By UAVs And Prior Map-guided Mobile Robot Navigation

The movement ability and operation mode of Unmanned Aerial Vehicles(UAVs)and ground mobile robots are different.UAVs can quickly perform large-scale environmental perception and map construction from an aerial perspective,but their endurance is limited.Ground robots have high loads and long endurance,but their ability to pass through complex scenes is limited.Therefore,the topological map construction by UAVs and prior map-guided mobile robot navigation are studied in this thesis.Aiming at the problem of UAV indoor mapping,this thesis proposes an indoor mapping method based on the truncated distance field and the sparse topological map.In the design of an indoor 3D reconstruction algorithm,the truncation distance is set to reduce the number of grids and the amount of calculation.To meet the needs of navigation,the Euclidean Signed Distance Field is incrementally constructed in real time in the dynamically growing Truncated Signed Distance Field.Aiming at the problems of large memory occupation and low transmission and multiplexing efficiency of maps,a 2D sparse topology map construction method based on Euclidean Signed Distance Field is proposed.The map only retains 2D nodes and connecting edges that can be extracted through space.To verify the effectiveness of the proposed indoor mapping method,experiments are carried out,such as simulation,dataset and real scene experiments.The comparison results with other indoor mapping methods show that the map constructed by the proposed method is lighter and has higher reuse efficiency.Aiming at the problem of UAV outdoor mapping,this thesis proposes an outdoor mapping method based on accelerated image-stitching and lightweight Res Net.Due to a large amount of computation and low efficiency of multi-image stitching,a parallel accelerated image stitching algorithm based on GPU is designed to achieve a consistent description of large-scale outdoor scenes.Aiming at the problem of high segmentation accuracy but low efficiency of a multilayer neural network,the segmentation speed is increased by a lightweight residual block,and the dilated convolution module is added to increase the receptive field to ensure segmentation accuracy.To make the outdoor map more effectively used to guide the mobile robot navigation,the topology map is constructed by using a skeleton extraction algorithm after the road segmentation.To verify the effectiveness of the proposed outdoor mapping method,the proposed method is tested on public datasets and real scenes.The comparison results with other methods show that the proposed method has a superior mapping effect.To realize the autonomous navigation of mobile robots guided by the prior topological map obtained by UAVs in indoor or outdoor scenes is one of the focus of this thesis.Firstly,the graph search algorithm is used to search the global path,and a safe and feasible global path connecting the starting point and the ending point is obtained.The shortest path obtained by the greedy algorithm is too close to the obstacle and has poor safety in navigation.Therefore,this thesis proposes a local obstacle avoidance method based on front-end kinematic trajectory search and back-end B-spline trajectory optimization.The method uses the Pontryagin minimum principle to search for a safe kinematic trajectory with the least cost,and initially achieves local obstacle avoidance at the front end.In addition,to optimize the local trajectory,the B-spline convex hull characteristic is used to constrain the trajectory to keep a safe distance between the trajectory and the obstacle.At the same time,considering the constraints of the movement ability of the mobile robot,the time distribution is adjusted by adjusting the node span,so that the robot can re-plan a safe,smooth and dynamic local trajectory.Test experiments are carried out in real scenarios,and the experimental results verify the effectiveness and practicability of the method proposed in this thesis.