Object Recognition Method And Grasp Planning Strategy Of Soft Robotic Gripper

In the unstructured environment,the soft gripper has wide application prospects like deep-sea exploration,medical service and family service.The object recognition and grasp planning of soft grippers play a decisive role in the achievement of successful grasping,and they are indispensable requirements of numerous application scenarios of the soft gripper.In order to make the object recognition and grasping more accurate and efficient,this thesis designs and manufactures a kind of soft gripper that has the function of contact force perception and bending perception,then proposes a method of the object recognition which combines the visual and tactile perception and studies the grasp planning strategies of the soft gripper.The main research achievements are as follows:1.The design and manufacture of soft grippers,and the establishment of visual and tactile sensing systems.The flexible force sensors and bend sensors are selected as tactile sensors,and a kind of soft gripper which has the function of contact force and bending perception is designed and manufactured.On this basis,the visual perception system and tactile perception system are established,which can acquire the tactile sequences,RGB images and depth images of objects laying a good foundation for the investigation on the methods of object recognition and grasp planning strategies of soft grippers.2.The method of object recognition combining visual and tactile perception.The two object detection algorithms,namely SSD and Faster RCNN,are used for visual recognition based on the visual perception system of the soft gripper.Based on the soft gripper and its tactile perception system,statistical features of the tactile sensor data are extracted as tactile features which are classified by using machine learning algorithms to realize the object tactile recognition.Based on the theory of multi-mode data fusion,a method of object recognition that combines the visual and tactile perception on the decisive level is proposed for the first time.The method of object recognition combining visual and tactile perception has a recognition accuracy of 98.7%,which is 9.94% higher than that with visual or tactile perception only.3.The grasp planning strategy of soft grippers.In terms of the grasping scene where objects are placed on the plane,the method of visual location and posture estimation of object is studied by means of camera calibration principle and point cloud matching algorithm.Drawing on grasp postures of human hands,three grasp modes of the three-finger soft gripper are generated,and a grasp planning strategy is put forward which is based on the geometric center of objects and feasible grasp modes,thus solving the problem of grasping objects of different shapes.According to the grasp planning strategy,the soft gripper can grasp the objects successfully,which verifies the feasibility and effectiveness of the grasp planning strategy.The research findings of this thesis are of great significance to the basic theories of perception,recognition and grasp planning of soft grippers,which is bound to promote the practical application of soft grippers.