Research On The Grasp Planning Method Based On Gaussian Process

Autonomous grasping technology has been widely considered to be a basic part of robotic intellectualization. In recent years, with the development of intelligent robot, the demand of autonomously capture is increasingly high. However, master-slave operation is mainly adopted in the robot technique, which is time-consuming and di?cult. The robot is controlled to grab object through operating lever, which requires operators passing through professional training. Therefore, it is vital to conduct the research on autonomous grasping. To this end, an algorithm of autonomous grasping based on Gaussian Process Classifier is proposed in this paper. On this basis, the operation of grabbing is completed through the underactuated mobile manipulator.The known and unknown grasping points of objects can be plan based on the grasping planning method in this paper. Planning unit acquires the environmental points cloud through the RGB-D sensors and the target object points cloud through plane extraction method. On the basis of that, the stability of scraping and the principle of force closure are combined to extract the five grabbing features. The training of the Gaussian Process Classifier is finished through the training set obtained by machine learning algorithms from introduction. Finally, grabbing features are taken as input for Gaussian Classifier,and some probabilities of feasible catching points are given. By repeating the above steps with random sampling method, feasible grab points for objects can be obtained.Design grabbing simulator and optimize grabbing path of the underactuated hand.A quasi-static model of the underactuated hand is established through the principle of virtual work. Furthermore, the underactuated hand grasping simulator is built in Matlab.The grabbing simulator takes the grasping points from planner as input and the minimum offset for grabbing points as the optimization goal to get the optimal grab path.The grab control strategy is designed based on contact detection. To begin with,Gaussian process regression is used to calibrate the relationship between the current and the rotating angles of underactuated units. Moreover, continuous detections for the current in the catching process are taken to judge whether the underactuated hand contact with the object. The whole process is divided into 3 stages of contacting, clamping and releasing to realize respective control.Establishing the platform of autonomous grasping system and experimental research are studied in this paper. Experiments are conducted for the grasping planning method,grasping path planner of the underactuated hand and contact detector. The feasibility of the various components is verified. The experiment platform and the software architecture of the system are built. Through the communication mechanism under the framework of ROS(robot operating system), the information interaction between subsystems is established to catch the complete experiment.