ICONIP2016 - Multiple Pregrasping Poses Prediction Using Combining Deep Convolutional Neural Network and Mixture Density Network
Multiple Pregrasping Poses Prediction Using Combining Deep Convolutional Neural Network and Mixture Density Network
Abstract.
In this paper, we propose a deep neural network to predictpregrasp poses of a 3D object. Specifically, a single RGB-D image is used to determine multiple 3D positions of three fingers which can provide suitable pregrasps for a known or an unknown object in various poses. Multiple pregrasping pose prediction is typically complex multi-valued functions where standard regression models fail. To this end, we proposed a deep neural network that contains a variant of traditional deep convolutional neural network, followed by a mixture density network. Additionally, to overcome the difficulty in learning with insufficient data for the first part of the proposed network we develop a supervised learning technique to pretrain the variant of convolutional neural network. abstract environment.
Keywords: Grasping pose prediction, deep convolutional neural network, mixture density network
