Research Of Object Perception And Grasp Control Based On Dexterous Hand Tactile

Perception of external environment is the premise of intelligent control and interaction of robot,and the basis of fine operation of robot.As an important sensing mode in robot system,tactile not only reflects the interaction between robot and environment,but also reflects the physical properties and spatial information of the contacted objects.In the past 20 years,with the rapid development of sensor technology,tactile sensor has gradually become the "standard configuration" of smart robot dexterous hand.It is of great significance to study how to acquire,understand and use the tactile information for improving the robot's environmental perception ability and fine operation of dexterous hand.Focusing on the research of object perception and grasp control method based on dexterous hand tactile,an experimental system consisting of Barrett BH8-282 dexterous hand,Schunk LWA 4D robot arm and Star Form Swift camera is set up.In this paper,we focus on the problems of object hardness recognition based on tactile,object geometry recognition based on visual and tactile point cloud,and dexterous hand autonomous grasping based on visual-tactile fusion.In order to recognize the object hardness based on dexterous hand tactile,a CNN-LSTM combined neural network model is proposed by analyzing the temporal and spatial characteristics of tactile data of dexterous hand.And the model training and verification are carried out with a self-built tactile dataset.The recognition accuracy of 95.0% is obtained.In addition,the PD grasp control based on object hardness online recognition is designed,realizing the flexible grasping of different hardness objects by dexterous hand.In order to perceive the geometric characteristics of the object,a method of 3D point cloud generation based on dexterous hand tactile is proposed.By designing a reasonable dexterous hand exploration program,the robot can collect 3D point cloud of fixed object with known position,and then recognize the shape,size and attitude information of the object.Secondly,this paper attempts to integrate the visual point cloud and the tactile point cloud,and reconstruct the model surface.With this method,more accurate recognition results of the geometric attributes of the object can be get.Aiming at the task of autonomous grasping of dexterous hand,based on the above research,this paper designs and realizes the autonomous grasping function of dexterous hand system based on visual-tactile fusion.In the process of autonomous grasping.Firstly,the position,attitude and size information of the target object is obtained by using the visual-tactile point cloud;Secondly,the system directly queries and adopts the pre stored dexterous hand posture according to the information.Then,the robot arm path planning algorithm is used to guide the dexterous hand to reach the target position.Finally,in the grasping execution stage,the grasping force is controlled precisely based on the tactile perception.The system is used to grab 14 kinds of objects,and the average success rate is 88.57%.