Catching Object in Flight Based on Trajectory Prediction on Camera Space

Abstract

In order to ensure the robot to perform a task of catching a moving object, we present a robot catching method based on the prediction of moving object's trajectory on camera space. In the proposed method, the method of camera-space manipulation (CSM) is employed to obtain the position information of the moving object under the three-dimensional space, and support vector regression (SVR) which learning the nonlinear dynamic of the object's movement from the object's historical movement data is adopted to predict the moving trajectory of the object, according to which the robot's motion is planned. The view parameters of CSM is updated with the motion of the robot thus gradually calibrating the information of moving object's three-dimensional space position. Then re-predict the trajectory and re-plan the robot. Loop for the above steps, the end-effector of the robot will move closer and closer to the moving object, and then the task of robot catching the moving object is realized. Experimental results performed in simulation have validated the effectiveness of the proposed method.