Research On Spatial Target Grasping Based On Vision For Manipulators

Spatial manipulators are able to substitute cosmonauts to accomplish manyimportant tasks, which makes it very worthy of study. When executing spatial tasks,the manipulators need to recognize, track and grasp the target. Vision is an easy accessto the environment information. A research into vision based spatial target graspingcan offer a way to examine algorithms relating to vision, control and calibration.Moreover, it can provide an experimental platform for spatial grasping.Firstly，to use machine vision, both intrinsic and extrinsic parameters of thecameras need to be calibrated. The calibration of intrinsic parameters is done with thehelp of MATLAB toolbox. Combining geometry method and pose-decompositionmethod, the calibration of extrinsic parameters of binocular vision is finished. Thecalibration of eye-in-hand system is sensitive to noise, and traditional calibrationmethod cannot meet the requirement of precision. In this paper, a non-linear optimaleye-in-hand calibration method is proposed, in which the rotation matrix andtranslation vector are calibrated simultaneously. This method makes the calibrationless sensitive to noise, and therefore more precise.Secondly, the pose of the target needs to be estimated. For a target characterizedby circle, the task can be transferred into circle pose estimation. If there are no specialmarks on the circle, it is hard to find the corresponding points in its projection. In thispaper, the global information of the projection is utilized to reconstruct the pose of thetarget. Because the projection of a circle is an ellipse, the expression of the ellipseneeds to be calculated before reconstruction. Based on least square method, the ellipseis well fitted. A closed form solution to estimating the pose of a circle based on itselliptical projection is deduced. When doing estimation, there are two possiblesolutions. An approach based on Euclidean angular constraint is used to overcome thisambiguity.Last, but not the least, the manipulator is controlled based on vision feedback. Toaccomplish the grasping task, the motion state of the target needs to be estimated. Formaneuvering spatial targets, kalman filter cannot well estimate their motion state. AnIMM estimator is implemented, which performs much better than kalman filter. Apartfrom this, because PBVS has no control on graph features, it is quite possible that thetarget may escape the field of view (FOV). An approach based on velocitycompensation for FOV is proposed. When the target is near the boundary of field ofview, the velocity of feature points in the projection is computed, based on which thevelocity compensation of camera is designed. With the help of MATLAB, the efficiency of this method is demonstrated.In this paper, a ground experimental platform of spatial manipulator isconstructed within the research background of spatial target grasping. Simulations andexperiments have been made to verify the algorithms.