Research On Grasping Position Estimation Algorithm Of Unknown Objects

As an important ability of service robots,the precondition which the manipulators could successfully grasp is to obtain the accurate grasping position of the target object.Several factors,such as the dynamic and uncertainty of the target environment,complex and varied target objects,impose higher requirements for the service robot grasping's intelligence.At the same time,as a target object without geometric model and feature information,unknown objects make the position of target objects more challenging for service robots.For the actual needs of service robots,we study the estimation of the grasping position of unknown objects based on vision.In order to realize the fast grasping position estimation of an object from a single perspective,this paper proposes a grasping position estimation method for an unknown object based on adaptive dynamic force balance,referring to the force balance characteristics in the grasping process.By using the principal axis search strategy,the position which can be used for grasping operation is obtained from the contour information of the target object,and the candidate grasping position is generated adaptively according to the concavity and convexity of the object.Combining with the force balance analysis of the object,the reasonable grasping position is estimated to improve the accuracy of grasping position estimation.In addition,this method is more universal without the need of offline training or building 3D point cloud models for objects.Aiming at the problem that inaccurate principal axis leads to lower accuracy of grasping position estimation,this paper combines the normal search strategy of neighborhood surface to obtain the normal vector set of multi-surface object accurately based on the original force balance analysis,and then combines the isotropic neighborhood segmentation algorithm based on Gauss mapping to find the data points and normal on the same surface.The force balance analysis is used to extract the optimal grasping position on each surface,and the grasping position of the unknown object is determined through comprehensive comparison,which can effectively solve the problem that the inaccurate principal axis leads to the failure of grasping position estimation,and improve the accuracy of grasping position estimation.Since the validity of the neighborhood surface normal search depends on the selection of the distance,the fitting residual and the normal deviation parameter.If the parameter is improperly selected,the number of iterations will be increased to compensate for the deficiency caused by the parameter being too large or too small.At the same time,each iteration needs to fit the plane,solve the eigenvector of the covariance matrix and the distance from the neighborhood point to the fitting plane,so the larger the number of iterations,the longer it takes.In this paper,an adaptive neighborhood surface normal search strategy is proposed.By introducing feature evaluation coefficients and fusing adaptive fuzzy C-means clustering algorithm,parameters such as distance,fitting residuals and normal deviation can be adaptively selected,thus improving the efficiency of the algorithm by about 15%.In order to verify the effectiveness of the grasping position estimation in this paper,we use KIT model base,YCB model base and Cornell database to obtain the grasping position set of objects by force-closed grasping simulation experiments,and make the grasping position evaluation database.Then we design a grasping position evaluation function based on the deviation measurement between grasping samples and standard samples.The experimental results show that for most common objects,the grasping positions estimated by the algorithm are within the error range(the max position deviation is 0.02m,the max axis angle deviation is 0.196rad).Meanwhile,the validity of the proposed algorithm is verified by the service robot grasping simulation system and the real manipulator grasping experiment.Finally,the research work of this paper is summarized,and the problems that need to be solved in the future are prospected.