Abstract

We propose a supervised end-to-end learning method that leverages on our grasp configuration network to predict 6 dimensional grasp configurations. Furthermore, we demonstrate a novel way of data collection using a generic teaching tool to obtain high-dimensional annotations for objects in 3D space. We have demonstrated more than 10,000 grasps for 7 types of objects and through our experiments, we show that our method is able to grasp these objects and propose a larger variety of configurations than other state-of-the-art methods.