Research On Obstacle Avoidance Of Quadruped Robot Based On Local Accessibility Map Construction

In the complex natural environment,foot-type mobility has a high adaptability to the environment,which makes the application of foot-type robots in field survey,disaster relief and other fields have a broad prospect.Nowadays,the application of multi-sensor fusion technology to improve the environmental adaptability of quadruped robot is the focus of scholars at home and abroad.In view of this,the depth camera with monocular color camera and IMU are used as the main sensors in this paper to complete the autonomous positioning and local map construction of the quadruped robot,design the terrain cost function,convert the generated local map into a passability map,and study the obstacle avoidance of the quadruped robot based on the RRT algorithm.In order to improve the passing ability of quadruped robot in the face of unknown and complex terrain environment.The main contents are as follows:(1)Establish the global pose description coordinate system of the quadruped robot,and establish the camera and IMU sensor model.Use Imu＿utils toolbox to calibrate white Gaussian noise and bias of IMU.Online calibration of external parameters between monocular depth camera and IMU is carried out.(2)After determining the type of feature points used in this paper,KLT optical flow method is used for tracking,which is used to estimate camera motion by pole-geometry method.IMU information is pre-integrated using the pre-integration method.In the VIO backend,the nonlinear information of vision and IMU is fused in a tightly coupled manner.The inverse depth measurement model is introduced and optimized by extended Kalman Filter(EKF)to obtain the pose estimation of the quadruped robot,that is,to complete the positioning work.(3)The transformation process of 3D point cloud from depth image is described.The voxel filtering method is used to reduce the size of point cloud to speed up the processing rate of point cloud.The raster height measurements and their variances,as well as the raster cell location uncertainty covariance are derived to obtain a local map with higher accuracy.The calculation formula of the estimated average height of each grid cell was derived,the map fusion process was completed,and the local map was constructed with quadruped robot as the center,that is,the environment reconstruction was completed.(4)The influence of terrain slope,roughness and concavity on the passing performance of quadruped robot was analyzed.The terrain features of grid were calculated by 3d point cloud data to design terrain cost function.Considering the motion performance of the quadruped robot,terrain cost function is used to transform the local map into the traversability map of the quadruped robot.The obstacle avoidance path of quadruped robot is planned based on RRT algorithm.(5)A ROS simulation platform has been built,including the motion control system,sensor system,robot positioning system,local map building system and robot obstacle avoidance system of quadruped robot.On the simulation platform,the localization experiment of visual inertial odometer,the local map construction experiment with robot as the center and the obstacle avoidance experiment of quadruped robot based on the local passability map are carried out.The simulation results confirm that the obstacle avoidance system of the quadruped robot designed in this paper based on the local passability map can effectively improve the passing ability of the quadruped robot when it passes through the terrain with obstacles.