Research On UAV Path Planning Algorithm Based On Visual Semantic SLAM

Simultaneous Localization and Mapping(SLAM)refers to the technology of achieving autonomous robot localization and mapping by using the robot’s own sensors to perceive the external environment,without relying on pre-existing maps or external positioning methods such as GPS.It is mainly used to solve the problem of localization navigation and map construction for robots in unknown environments.Given its important theoretical and practical significance,numerous scholars have recognized autonomous navigation as the pivotal element for enabling fully autonomous robots.Unmanned aerial vehicles(UAVs)have practical applications in fields such as drone inspection and logistics transportation due to their small size,high maneuverability,and other advantages.These tasks require drones to achieve realtime positioning,detect obstacles in the environment,and autonomously avoid them.However,as a trade-off for their light payloads,it remains a formidable challenge to accomplish autonomous positioning and path planning with limited computing power.With the development of deep learning,deep learning is widely used in many fields such as image detection and classification,natural language processing and big data analysis.Therefore,the integration of deep learning and SLAM algorithms has emerged as a research hotspot for enhancing the accuracy and robustness of SLAM algorithms.This thesis focuses on issues of positioning errors and path planning failures in traditional autonomous navigation methods for UAVs.The specific contributions of this work are as follows:(1)Introducing deep learning methods into the front of SLAM algorithms,a deep learningbased visual-inertial odometry(VIO)is designed to extract and compute features from sensor data using deep learning methods.The VIO is combined with the VINS algorithm,and its positioning accuracy is tested through experiments.The result shows that the method designed in this thesis effectively improves the accuracy and robustness of the SLAM algorithm.(2)In terms of safety and real-time performance of path planning algorithms,a path planning algorithm is proposed that integrates the advantages of both A*and artificial potential field(APF)methods.This improves the efficiency of path generation and reduces the possibility of path planning failure due to local optima.(3)The performance of the proposed algorithms is tested through experiments,including simulation tests using current open-source software for SLAM and path planning algorithms,as well as tests in a real-world environment using a UAV experimental platform and testing scenarios.The experimental results show that the proposed method effectively improves the safety and efficiency of task execution.