Dynamic grasping parameter estimation based on feedforward radial basis function network

Abstract

In this paper, we propose a motion estimation method based on a feed-forward radial basis network for grasping arbitrary moving objects. We first employ a kernel correlation filtering (KCF) algorithm to track the target position in real-time and establish the motion model of the target. Using the feed-forward radial base network, we then adjust the sampling time of the Kalman filter (KF) to predict the motion parameters of the target. Since that, we can reduce the computing time and improve the accuracy of the estimation of the motion parameters. Compared with the feed-forward perceptron network, the proposed method shortens the required time for grasping by 20%, which can avoid a failure grasp due to the arbitrary movement of the object in grasping.