Off-Road Environmental Traversability Analyzing For Quadruped Robots Based On Depth Sensing Information

Today,with the rapid development of unmanned technology,as an important intelligent unmanned system,robots are widely used in agriculture,industry and national defense,and have important significance in environmental detection,material transportation and other applications.Among mobile robots,they are mainly divided into wheeled robots,leg-footed robots and crawler robots.Among them,the quadruped bionic robot has flexible motion control and special traveling characteristics,has good adaptability to the unstructured environment in the field,and is widely used in the field environment.With the increasing social requirements for robot intelligence,the autonomous walking of quadruped robots has gradually become a research hotspot.To realize robot autonomous walking,environment awareness is an indispensable important technology.Terrain traversability analysis technology as part of environment awareness technology is an important basis for robot autonomous walk.In this topic,in response to the task requirements of the quadruped robot autonomous walking,the environment perception system of the quadruped robot was built based on the Realsense D435i depth camera,and research was conducted on the traversability of the quadruped robot in an unstructured environment.The main research contents are as follows:(1)Build a quadruped robot environment perception system and establish a terrain recognition database.The depth camera uses the Kalibr calibration method to calibrate the internal parameters of the camera.Use the vision system to collect color pictures of various terrain types(such as:grassland,mud,stone brick road,asphalt road,snow,etc.).The terrain classification data set is established,and the image of the data set is preprocessed by using image filtering technology and histogram equalization technology.(2)Construct a 2.5D elevation grid map model centered on a quadruped robot,and extract the geometric features of the terrain in the grid.Calculate the geometric features(slope value,curvature value,roughness,undulation,etc.)in the grid through the dense point cloud information obtained by the depth camera to describe the terrain in the grid.At the same time,an improved method for calculating the roughness characteristics is proposed to improve the calculation accuracy.(3)Based on the support vector machine to identify and classify the terrain types,and combine geometric features and terrain types,a terrain traversability analysis method suitable for quadruped robots is designed.The gray level co-occurrence matrix,direction gradient histogram and color distribution histogram are used to describe the picture features,and the features are linearly fused.A support vector machine model with a "one-to-one" structure is used to learn and train the terrain features,so as to obtain a terrain classification model.Aiming at the grid terrain,the geometric features are calculated and the terrain types are recognized at the same time.The calculation results are combined and the robot structure is used as the basis to design the terrain traversability parameters,thereby improving the accuracy of the traversability.