Research On The Key Technology Of Target Grasping For Intelligent Robot

As more and more intelligent robots used in various fields,and grasping operation is a research hotspot in the field of robots,the realization of intelligent grasping is of great significance to the rapid development and commercial landing of robots.At present,there is still a certain gap between the research and the realization of real intelligent robot grasping.There are three main problems.(1)An accurate system model needs to be established before the robot grasping operation.Many studies ignored the positioning error of the manipulator,which makes the actual position of the robot different from the desired position.(2)The traditional robot target grasping methods require accurate geometric model information of the object,which can't be used in the unstructured scenes.The autonomous detection,searching and grasping of the target object in a variety of unknown scenes is the biggest challenge of the current intelligent robot grasping research.(3)Due to the cost and sensitivity limitations of fingertip tactile / force sensors,many robot hands currently do not process the capability of tactile / force-sensing.Many objects in the actual environment,especially soft and fragile objects,have very high demand for the grasping force.Excessive grasping force will damage the objects.In order to enable intelligent robots to grasp objects in unstructured environments accurately,autonomously and safely,and increase the types of objects that can be grasped widely,this paper has carried out three key technologies for intelligent robot target grasping corresponding to the above three problems.The three key technologies include robot kinematics and vision system calibration methods,vision-based intelligent robot target detection and grasping methods,tactile-based robot compliant grasping control methods.Firstly,in order to improve the positioning accuracy of the intelligent robot when the robot grasping objects,a low-cost and easy-to-operate kinematic calibration method for manipulator based on straight-line virtual constraints is proposed.The robot kinematics model is established using the improved DH method.Then the kinematics error model based on the straight-line constraints is established.An experimental system for kinematic calibration of the manipulator is designed.The experimental results of the kinematic calibration show that the algorithm improves the positioning accuracy of the manipulator effectively.In addition,the robot vision system modeling and calibration methods are studied and implemented,including camera modeling and calibration,“Eye-to-Hand” system calibration and “Eye-in-Hand” system calibration.The accurate robot vision system model is established,and the relationship matrices between different coordinates in the robot vision system are obtained.Secondly,in order to make intelligent robots detect and grasp target objects autonomously in unstructured actual environments,a vision-based intelligent robot object grasping system is designed.The object detection algorithm based on RGB images is studied,and real-time detection to the RGB images streaming captured by camera is realized.The registration method of depth image and RGB image and the inpaint method of depth image are studied and implemented,which improved the accuracy of depth information.In addition,a rotation-free target grasping algorithm based on the detection constraints is proposed.A grasping pose estimation algorithm is designed to estimate the grasp pose of the target object from the object detection results directly.The detection and grasping operation of various types of target objects is successfully implemented in the unstructured actual environments.Thirdly,due to the target object may be partially or completely occluded in the unknown scenes,the robot needs to be able to detect and grasp the target object and be able to grasp or remove the unrecognized objects.A target detection and grasping algorithm considering rotation is proposed.The grasping is represented by the two-point representation,and the rotational angle of the robot is considered.The grasping poses generation algorithm based on the convolutional neural network is studied.The grasping poses are generated quickly and efficiently for the input depth image.Then the target detection and grasping algorithm structure in the unknown scenes is designed.Experimental results of robot target grasping show that the algorithm can search and grasp target object in a variety of unknown scenes effectively.Compared to the rotation-free target grasping algorithm based on detection constraints,the target grasping success rate is further improved.Finally,in order to achieve a safe and stable grasping operation of objects with different characteristics by intelligent robots,especially soft objects and fragile objects,a new flexible electronic skin tactile sensor based on hydrogel materials was designed.This tactile sensor has a sensitive force-sensing capability,which is low cost,convenient to use,and can be directly worn on a robot to measure the contact information between the robot and the environment.The compliant grasping algorithm of the robot hand based on the tactile sensor is studied.The grasping force is controlled by the compliant algorithm when the robot grasping the object.The tactile sensor is integrated on the Kinova KG-3 robot which can only perform position control.A compliant grasping control experiment of the robot based on tactile perception is carried out,which makes the robot enable to grasp soft and fragile objects.The safe grasping operation of fragile and easy-to-damaged objects is successfully realized,such as grape,tofu,quail eggs and potato chips.