Research On SLAM Algorithm Of Legged Robots

With the development of science and technology,robot-related industries are applied to all aspects of human life.The increasingly complex task requirements put forward higher requirements for the flexibility and reliability of robot hardware structure,and the intelligence and robustness of software algorithms,and as a bionic robot,the legged robot is more mobile than the wheeled robot and the pipeline robot,and has better adaptability to complex terrain.Robot autonomous navigation technology is an important research direction in the field of robot research,and SLAM is one of the key technology in robot autonomous navigation technology.This paper takes the Visual SLAM algorithm as the main research object,combined with the actual working scene of the legged robot,researches and implements a Visual SLAM method based on the Manhattan World assumption.The main work of this paper is as follows:1.Manhattan World building algorithm based on structured features.The line feature is introduced in the visual front-end,which improves the performance of Visual SLAM in the environment of texture loss and image blur,and improves the extraction and matching algorithm of point and line features;Proposed a line-based Manhattan World-building algorithm,implemented visual-inertial based Manhattan World detection and tracking algorithm,and optimized the real-time and robustness of the Manhattan World construction algorithm in the visual front end.2.A Visual odometry algorithm for legged robots based on the Manhattan Wor-ld assumption.Solved the problem of VIO scale estimation on legged robots through stereo vision,and designed a visual-inertial odometry algorithm based on point-line feature fusion;constructed line feature geometric constraints based on the Manhattan World assumption,and corrected pose of legged robot by visually estimating the direction of gravity.3.Pose correction algorithm for legged robots based on structured features.A loop closure detection algorithm based on HF-Net is designed,and a Manhattan World constraint construction algorithm based on local map is implemented according to the Manhattan World assumption,and a pose graph optimization algorithm for multi-source constraint fusion is designed.4.Integrating the above algorithms,a kind of Visual SLAM algorithms for legged robots was designed and implemented,named Manhattan-SLAM.The software and hardware platforms of the legged robot experiment platform were built,and the accuracy and robustness verification experiments were designed to verify the feasibility and effectiveness of the relevant algorithms in the legged robot.