Tactile Stability Analysis And Classification For Robot Fine Operation

There are three main ways for human feedback: tactile feedback,visual feedback and acoustic feedback.The data collected from the tactile sensing system is an important medium for environmental exploration and identification no matter for human or robots,and it is also a necessary condition for human-machine interaction.For now,the types of data collected on robots are roughly divided into: tactile,visual,depth,various joint angles,etc.For example: dexterous hand finger joint angle,wrist angle and so on.These data can be used alone as a guide for fine operations or can be used as a fusion modality to guide the robot to complete the fine manipulation tasks.In tactile classification,by the manipulator attached to the robot actuator and the process which object contacted with,the joint sensor records the changes of each joint angle in the crawling process,the tactile sensor records the force changes of the object during the entire grasping process and then send these tactile time series and joint angles into the terminal for preservation which compensates for the lack rely on laser radar,wrist force sensor and distance sensors.Through processing the collected rich data and sending them into different classifiers can effectively improve the robot’s grasp stability analysis in complex environment and accurate classification.Thus to improve the robot’s perception of the surroundings.Firstly,this paper introduces the research status of grasping stability analysis and tactile information classification in the field of robot intelligent control,the advantages and disadvantages of using the potential energy method to analyze the stability of object grasping are enumerated through the literature of other researchers,therefor,this paper presents the rationality and innovation of using tactile information and robotic joint angle to analyze the grasping stability.Then,the idea of using kernel is put forward to model the tactile information and construct the DTW kernel and GA kernel with joint angle of the robot,then send them into different machine learning for processing and analyzing.By analyzing the experimental results we can see that the experimental results of GA kernel perform better than DTW kernel,and theresults of the Extreme learning machines are better than Support vector machines.Secondly,the DTW kernel and GA kernel model are built using the existing tactile adjective data sets.And then send them into the machine learning for adjective attribute classification.In order to compare the result of the classification algorithm,the nearest neighbor algorithm is applied for object recognition.These conditions of all experiments are the same in order to maintain the same comparability of the results.Different algorithms have their own advantages and disadvantages,but it is obviously that the results of the nearest neighbor algorithm perform worse than using Extreme learning machine and Support vector machine,besides,the results of GA kernel perform better than DTW kernel.In our future works,this kind of algorithm will be used to real-time operaction that the object can be identified and recognized in the process of manipulating.