Object Recognition And Classification Method For Anthropomorphic Hand Based On Pneumatic Tactile

With the improvement of robot intelligence and automation,robot technology has been widely used in military medical,remote sensing control and life services.It is a key task for the robot to recognize the object correctly according to the sensor,which is very important for the realization of human-computer interaction and fine operation.The development of traditional visual recognition technology is relatively mature,which can effectively classify the object according to the contour shape of the detected object,but its recognition accuracy is easily affected by external environmental factors such as light or occlusion.Tactile information is another important modal information in robot perception system.It can sense the physical properties of objects such as material,hardness and shape through direct contact.The object recognition method based on tactile can make up for the lack of visual recognition and classification technology.The two key steps of robot tactile recognition classification method are to build an effective tactile data set and build a high-performance classification model,which determines the final classification accuracy of the recognition method.Therefore,this paper studies the underactuated rigid soft anthropomorphic hand experimental platform,and the specific research contents and results are as follows:On the basis of studying the working principle of MPL115A2 barometer,a pneumatic tactile array sensor was fabricated by pouring silica gel,and a tactile information acquisition system was designed based on Kalman filtering algorithm.With the purpose of verifing whether the acquisition system can meet the requirements of robot identification and classification,performance evaluation experiments are carried out.Experiments show that the acquisition system can effectively sense the change of tactile information in the process of grasping,and the sensor has the characteristics of high sensitivity and small repeatability error.When the robot grasps the object to build the tactile data set,in addition to the performance of the tactile acquisition system,it must also ensure that the manipulator can grasp the object with accurate force.Because the flexible knuckles will deform according to the shape of the object when grasping the object,which has great uncertainty,the traditional modeling method can not be used for force tracking control.Therefore,this paper designs a force control system based on Fuzzy PI technology,and carries out the relevant hardware selection and debugging,completes the construction of the control system,which can effectively track the fingertip pressure.The accuracy and classification efficiency of recognition algorithm directly affect the direction of robot decision-making and the fluency of action.Traditional ensemble learning algorithms,such as random forest,do not change the classification model once it is constructed,which is lack of flexibility and low recognition accuracy in some scenes.In order to make the classification model dynamically adjust according to the test cases,this paper designs an improved random forest algorithm based on KNN algorithm.In addition,PCA technology is used to reduce the redundant dimension of tactile features,so as to reduce the time cost of data processing and improve the classification efficiency of the algorithm model.Combined with tactile information acquisition system and force control system,this paper builds an experimental platform of the underactuated rigid soft anthropomorphic hand.By grabbing objects to record the pressure change of fingertips and build the data set of the tactile information of palms,and the classification experiment is carried out according to the improved random forest model.The experimental results show that the fuzzy PI control technology can track fingertip pressure better,the improved random forest algorithm has higher recognition accuracy and classification efficiency than the traditional algorithm,and the anthropomorphic hand grasping recognition system can effectively complete recognition and classification based on touch.