| Good autonomous mobility of the rovers is a necessary condition for planetary exploration missions.The complex environment on the surface of the planet will cause geometric and non-geometric risks to autonomous mobility of the rovers.The geometric risks mainly refer to gian t rocks and steep slopes,which can be easily detected by the topographic elevation information.Correspondingly,the non-geometric risks are mainly caused by the interaction between wheels and terrain,such as rover sinking into soil,etc.,which is relatively difficult to detect.The qualitative classification of the terrain based on vision can help the planetary rover to identify the non-geometric risks to some extent,but this method lacks the quantitative evaluation of the mechanical properties of the terrain.Both wheel-terrain interaction classes and models are different for rovers moving on various terrain types.Therefore,when analyzing the wheel-terrain interaction forces and estimating the mechanical property parameters of the terrain,the corresponding wheel-terrain interaction model should be selected according to the actual working terrain environment.In order to solve the above problems,this paper studies the identification of the terrain mechanical properties of the contact area and the est imation of the terrain mechanical properties of the proposed contact area.This paper analyzes the gray variation of images for different terrain types,and then proposes multi-scale gray gradient level features,multi-scale edge-strength level features,multi-scale mean amplitude features in frequency domain,multi-scale amplitude-moment features in frequency domain and multi-scale spectrum-symmetry features of planet terrain surface.The texture features of the gray level co-occurrence matrix,the Gabor texture features,and the statistical features in space domain are also extracted from terrain images.All those features are used to construct the feature vector of planet terrain surface textures.Based on the feature vector,four classification algorithm s are adopted for terrain classification.The results of Mars terrain classification show that the classification accuracy for the feature vector of planet terrain surface textures is higher than that for feature vector of classic image texture s.In addition,the classification accuracy of the random forest is the highest among the four classification algorithms,reaching 94.7%.The wheel-terrain interaction models are different for a rover moves on different types of terrains.In order to select wheel-terrain interaction models,the wheel-terrain interactions are divided into three classes: two supporting lugs being adjacent,two supporting lugs may not be adjacent,and lugs entering into soil.The vibration models of a wheel under three classes of wheel-terrain interaction are built.The characteristics of wheel vibration under each class of wheel-terrain interaction are revealed.Four non-speed coupled features of the wheel vibration are extracted from the vibration model s of the wheel.They are used to construct the feature space of wheel vibration.According to the boundary conditions of three wheel-terrain interaction classes,the feature space domains of wheel-terrain interaction classes are analyzed.A method for recognizing wheel-terrain interaction classes is proposed based on their feature space domains.Experiments results show that the recognition accuracy of the wheel-terrain interaction classes reaches 92%.It laid the foundation for selecting wheel-terrain interaction models.This paper proposes a method for selecting wheel-terrain interaction models based on the recognition of wheel-terrain interaction classes.Based on the model selection,the method for identifying mechanical property parameters of various terrain types is proposed.Wheel slip ratio and sinkage are necessary for parameter identification.The model of slip ratio estimation is built based on wheel-terrain images.Wheel forward displacement and wheel rotation angle are detected from wheel-terrain images.They are input into the model of slip ratio estimation for calculating wheel slip ratio.The error of slip ratio estimation is proved to be less than 9% by experimental validation.The wheel sinkage for rovers traversing rough terrain is defined.A model for estimating wheel sinkag e is built.Then wheel sinkage is detected using wheel-terrain images.Experiment results of wheel sinkage detection prove that the method has high accuracy and strong adaptability of light.Both the detection methods of wheel slip ratio and sinkage are applied to a planetary rover experiment prototype.Then the planetary rover experiment prototype identifies mechanical property parameters of hard terrain,sandy terrain and gravel terrain.The surface texture features and mechanical property parameters of various terrain types are used to construct estimation system of proposed contact terrain mechanical properties.The terrain property database and the reasoning mechanism of terrain mechanical properties are built.They help the rover deduce mechanical properties of the proposed contact terrain according to its surface texture features.Then the spatial distribution of terrain mechanical properties is analyzed.In addition,a self-learning method of this system is proposed.The method enables the system to identify new terrain types.Then system will be updated to increase the texture information and mechanical property information of the corresponding terrain types.The rover's experimental system is built to verify the self-learning ability for the estimation system of proposed contact terrain mechanical properties.In addition,the ability for estimating mechanical properties of proposed contact terrain is also tested.This paper proposes the method for estimating the terrain mechanical properties of both contacting area and proposed contact area during the travel of a rover.This provided the basis for rover's high-fidelity simulation and the path planning with considering the terramechanics. |
PDF Full Text Request