Research On The Grasp Control Technology Of Dexterous Hand Based On Touch

With the continuous development of machinery and modern electronic technology,people have produced a variety of dexterous hands.It is an important medium for human-machine interaction between robots and the external environment.Therefore,dexterous hands have always been a hot topic in the field of robot.Compared with the structured industrial environment,the actual environment of service robot is more complex and diverse,with great uncertainty.Therefore,higher requirements are put forward for the robot's perception ability,which requires the robot dexterous hand to be able to perceive the environment through the sensor,and process and understand the sensing information.In this paper,the extraction and stable grasping of tactile information are studied.Tactile sensor is an important means for robot to perceive real environment.When the robot operates the object,the tactile sensor information can be used to judge the real state of the object.First of all,in order to solve the problem of grasping unknown objects without using visual sensors,a new tactile spatiotemporal feature vector which can be obtained with less computing resources is proposed based on the characteristics of tactile data time series.Then,in order to make the dexterous hand grasp the target stably,not only the correct grasping force is needed,but also the grasping position needs to be planned in advance.Therefore,the grasp planning of dexterous hand is needed.According to the D-H model of Barrett's dexterous hand,the kinematic model is established and the required joint torsion moment is obtained.The grasping coordinates of ellipsoid are established,and the grasping matrix of ellipsoid is derived.Using the performance index based on the grabbing matrix to get the best grabbing position,using the linear gradient flow algorithm to get the best grabbing force of ellipsoid.At the same time,the simulation results verify the correctness of ellipsoid grabbing program,and briefly introduce the boundary conditions of different structures of dexterous hand grabbing.Next,combining the temporal and spatial characteristics of the tactile,we construct a grabbing strategy based on the combination of multi support vector machine(MS VM)algorithm and k-nearest-neighbor(KNN)algorithm,and realize the efficient online recognition of unknown target state based on tactile sensor only.Finally,combined with the method proposed in this paper,the project classification experiment and grab experiment are carried out,and good results are achieved,which verify the effectiveness of this method.