Multi-tactile Fusion Perception And Grasping Control Of A Soft Robotic Hand

Compared with traditional rigid hands,soft robotic hands are made of soft materials and show high flexibility and safety,which are respected to be broadly used in materials sorting,fruit and vegetable picking and other occasions.In order to realize the adaptive grasping and recognition of objects by the soft hand,it is required to equip the soft hand with tactile sensing capabilities.The current soft tactile sensing and grasping systems are restricted by single sensing form and simple control methods,so they are difficult to adapt to complex and changeable grasping objects.To this end,this thesis proposes a novel soft hand with multi-tactile sensing capabilities from fingers and a palm,and conducts in-depth research on its multi-tactile fusion sensing methods and grasping control.The main contents of this article include:The integrated design of the soft hand and the multi-tactile sensing system.A novel design of the soft hand with pneumatically deformable fingers and palm is proposed,and a piezoresistive tactile sensor Flex Sensor 2.2''(abbr.Flex sensor)and a visio-tactile sensor Tac Tip are integrated into the soft fingers and the soft palm,respectively.The fabrication of integrating the soft structure and the sensor is introduced.Combined with the system configuration,the actuation and control system of the soft hand and the tactile sensing system are built.Research on tactile sensing algorithms of the soft finger and the soft palm.For the Flex sensor integrated in each soft finger,the sensing of bending angle of each soft finger is realized through circuit analysis and signal calibration.For the Tac Tip sensor integrated in the soft palm,a 2D contact region perception algorithm is proposed by image processing.And a 3D reconstruction algorithm for the deformed surface of the soft palm is proposed based on the perspective projection model of the camera.A sensing experiment for the soft palm was conducted to verify the accuracy of the tactile sensing algorithm.Research on multi-tactile fusion perception methods for object classification.Considering machine learning methods,a dataset of daily objects was established.Through the analysis of the tactile signals of the fingers and the palm,the spatial features describing the geometric information of objects are extracted.The tactile features of the fingers and the palm were fused,and the classifiers based on random forest(RF)and support vector machine(SVM)were designed and trained,respectively.The performance of the two classifiers was compared.The results showed that the performance of the SVM classifier was better,reaching an accuracy of 87.50% in the ten-class task in this thesis.Research on grasping control of the soft hand based on tactile sensing information.The robot grasping control system was built.On this platform,the research of the grasping control of the soft hand was conducted.First,based on the contact state perception of the soft palm,the adaptive grasping control of the soft hand was realized.Then,based on the optimization of the grasping configuration of the soft hand,the stable grasping control of the soft hand was realized.The effectiveness of the two grasping control methods were verified in the robot grasping system experiment.The research results of this thesis propose a solution based on the multi-tactile perception of the soft hand for intelligent grasping tasks and object recognition tasks,which is expected to be applied in the field of robot material sorting.