Simulation Research on Semantic Map Construction and Path Planning of UAV Based on AirSim

In order to meet the task requirements of autonomous navigation of UAV in the building space environment, the 3D dense mapping and path planning algorithm module of UAV is deeply studied. Aiming at the problem that ORBSLAM2 algorithm based on feature point method cannot build dense maps, dense point cloud maps and octree maps (Octo-map) are constructed incrementally. Through semantic segmentation and semantic information association, the construction of semantic Octo-map is realized. In response to the problem of multiple inflection points and obvious jitter in the path generated by the RRT algorithm, the smoothness of the path is improved by using quasi uniform cubic B-spline curves. Finally, the control scheme of UAV path tracking based on PX4 is designed, and the simulation of UAV path tracking in AirSim verifies the feasibility of path planning, which provides an effective simulation solution for the application of autonomous navigation technology of UAV.