SLAM Algorithm Of Quadruped Robot Based On Multi-sensor

The state estimation and localization of a quadruped robot are prerequisites for subsequent autonomous navigation,path planning,perceptual obstacle avoidance,and other high-level tasks,and can serve as feedback loops for the motion control system.Given the special legged locomotion mode of quadruped robots,the accumulated error of the inertial measurement unit(IMU),as well as the complexity of external sensor calculations,this study focuses on the state estimation and localization problems of a multi-sensor quadruped robot.The IMU measurement model,legged forward kinematics modeling,and visual simultaneous localization and mapping(SLAM)algorithm are utilized,and an invariant extended Kalman filter is employed to achieve accurate state estimation and localization of the quadruped robot.The specific research contents are as follows:Firstly,an overview is given of the quadruped robot and its sensors used in this study.For the task of estimating the body state of the quadruped robot,the attitude of the quadruped robot body is estimated by solving for the acceleration and gyroscope measurements from the IMU.Using the fourth-order Runge-Kutta algorithm,the position and velocity of the quadruped robot body are estimated by integrating the differential equations of the inertial navigation position and velocity.Secondly,considering the problems of insufficient feature point extraction and low localization accuracy of feature-based SLAM algorithms in low-texture environments,a structure-constrained point-line feature fusion SLAM algorithm is proposed.After extracting lines using the LSD algorithm,the lines are divided into parallel lines and perpendicular lines,which serve as constraints to optimize the 3D position and camera pose of the feature lines.The Manhattan principal direction of the environment is estimated using the CAPE algorithm,and it is used as a constraint to optimize the position and camera pose of the local map features,which improves the localization accuracy in weak-texture environments.Finally,to address issues such as large single-sensor errors,information loss,and perception delays,a multi-sensor fusion framework for quadruped robots based on an invariant extended Kalman filter is proposed.By representing the state vector as a Lie group that satisfies the invariant affine group property,and using invariant error theory and observation theory,the invariant extended Kalman filter eliminates the problems of positive feedback and inconsistency.The IMU prediction information,legged forward kinematics,and camera observation information are fused using the invariant extended Kalman filter to solve the problem of low single-sensor state estimation accuracy.