Research On Haptic Recognition Algorithm For Material Surface Texture

Haptic recognition is the frontier issue of robot intelligence,where the tactile classification and recognition of material surface texture is the prerequisite and guarantee for robots to be able to be autonomous.The robot can adjust its control strategy to grasp or manipulate the target by mastering the material properties of the object surface in the surrounding environment,thus preventing damage to the object.The material properties of the object surface also provide a key basis for the robot to choose the appropriate motion path or contact method,avoiding the hazards caused by potential risks to itself.The robot can acquire material information on the surface of an object through sensors with multiple sensory channels.Compared to visual and auditory information,the tactile information acquisition method is less affected by the working environment and more adaptable.When the robot works in unstructured environment,it will encounter common working conditions such as light changes and object occlusion,but the tactile recognition of material surface texture can still maintain good performance.In this paper,we study the feature extraction,data volume selection,signal similarity calculation and recognition robustness in the tactile recognition of material surface texture with the basic processing flow of recognition algorithm as the object,and provide reference for the development of the application of surface texture tactile information recognition,storage and transmission.The innovation points are as follows.(1)To address the problem of feature extraction of material surface texture haptic information,a joint feature is proposed based on the influence of texture structure and scanning behavior on the material surface texture haptic signal,including the normalized amplitude spectrum of surface texture haptic vibration acceleration signal and the statistical values of haptic scanning parameters,which simplifies the dimensionality of feature description.The singular spectrum analysis of the normalized amplitude spectrum verifies that the extracted features have good stability.Using the extracted features,a method of generating material surface texture haptic signals is proposed to alleviate the material surface texture haptic signal measurement limitations,improve the efficiency of material surface texture haptic signal acquisition,and verify the effectiveness of the joint features.(2)To address the boundary problem of available data length in material surface texture haptic recognition,the proposed two indicators of feature information gain and feature cohesion-separation coefficient of material surface texture haptic data are used to clarify the impact of data amount selection of material surface texture haptic information on recognition performance.In addition,based on the machine learning algorithm and the extracted joint features,the material surface texture is recognized with an accuracy of 93%,and the recognition method is verified to have good generalization capability using hypothesis testing.(3)To address the problem of calculating the similarity of tactile signals of material surface textures,a neuronal simulation model of tactile mechanoreceptors is used to transform tactile vibration stimuli into pulse sequences,and an interpretable mapping relationship between objective measures of vibrotactile similarity and subjective perception is established,which reduces the error between objective calculation and subjective evaluation of tactile signal similarity of material surface textures.Using the proposed similarity calculation method,the similarity between multiple material surface texture tactile signals in the public data set is calculated,and the feasibility of the neuromorphic approach in material surface texture tactile recognition is verified.(4)To address the problem of robustness of material surface texture tactile recognition,some neuromorphic models for converting material surface texture tactile signals,including vibration acceleration,contact force and scanning speed,into pulse sequences are proposed based on the physiological response characteristics of various tactile mechanoreceptors and the process of finger-surface interaction,and a neuromorphic method for material surface texture tactile recognition is proposed based on these models.Compared with the joint feature-based recognition,the recognition performance in the low signal-to-noise ratio case is significantly improved.In addition,the results can be used to guide the recognition of multiple tactile information fusion.