Research On Object Touch Recognition And Adaptive Grasping By Robot Dexterous Hand

Grasping unknown objects is the basic ability of robotic dexterous hand to perform various complicated tasks.The retrieval of unknown objects and the reconstruction of the model are preconditions for the robot to complete the grasping and manipulation planning.At present,robotic vision system is widely used in the extraction of object information.However,in non-structural environment,such as complex background,transparent objects,occlusion and darkness,the visual sensor may experience noise,resolution decline and even information loss,resulting in the failure of the grasping task.The paper aims to improve the autonomy,stability and self-adaptability of dexterous hand grasping in case of blind environment,this paper aims to complete model reconstruction and grasping through the multi-touch and multi-sensor information advantages of dexterous hand.The main research contents include unknown object touch exploration method of multi-finger dexterous hand,unknown object model reconstruction based on tactile information,grasping points optimization based on tactile information,and adaptive grasping method for object pose error.In order to realize the autonomous acquisition of unknown objects tactile information in the environment of blind information,an observation experiment on human touch exploration is made.Based on the observation and analysis of the exploration behavior of human for the unknown objects by tactile sensation,a method suitable for robots to conduct autonomous exploration of unknown objects by the tactile sense is proposed in this paper.According to human explo ration experience and object geometry,the haptic exploration process is divided into two phases;in the first stage,the top surface of the object is explored,and according to the top tactile information,the basic object size and posture are estimated,and bounding box method is used to approximate objects;in the second stage,the objects are classified according to the basic dimensions based on the classification discrimination inequality,and different side exploration strategies are designed for different objects to explore the sides of the objects.An autonomous exploration grasping system based on humanoid manipulator and multi-fingered dexterous hand is built for verification.The experimental results show that the robot can explore unknown objects by tactile sense in any size and shape autonomously by using this method,meanwhile avoiding the collision between the robot and the unknown objects,so as to lay the foundation for the model reconstruction.A reconstructing method for the unknown object model based on tactile location and normal vector information is proposed for tactile point cloud data obtained from tactile exploration.The clustering algorithm is used to classify the point cloud,of which the geometrical features are then extracted,to construct the geometric feature vector.The object shape parameters are identified according to the feature vector and the intelligent classification algorithm.Then,the 2D shape descriptor of the object is constructed and the pose and size parameters ar e obtained by matching algorithm.Finally,the unknown object model by the super quadric surface equation is constructed.Experimental results show that the proposed method has high robustness to noise,requires less data and has high accuracy of model reconstruction.Based on the multi-touch of the dexterous hand and the information of the known tactile exploration,the grasping point optimization method is studied.According to the grasp stability index,the initial grasping point and the robot motion parameters are optimized in the known information point set to simplify the grasp planning and improve the real-time performance.Under the constraint of dexterous hand kinematics and manipulability,the optimal grasping points in the search region are obtained by the KNN-based optimization search algorithm,and the effectiveness of the method is verified by simulation and experiment.As the objects are prone to the grasp failure caused by pose error in the blind environment,the adaptive grasping method based on force collaborative grasping and error compensation is proposed to solve the problem.The grasping process of touching object successively and conducting grasping force simultaneously is completed by the force collaborative grasping based on impedance control and the force control,to improve the stability of the object in the grasping.The position and posture error of the object is estimated by fingertip tactile information and the stability of grasp is analyzed.The finger or arm of robot is applied for the adjustment to compensate the error for different ranges of pose errors.The experimental results show that the adaptive grasp method can effectively improve the success rate of grasping objects for dexterous hand in the blind environment with the pose error.