Deep-LfD: Deep robot learning from demonstrations

Abstract

Like other robot learning from demonstration (LfD) approaches, deep-LfD builds a task model from sample demonstrations. However, unlike conventional LfD, the deep-LfD model learns the relation between high dimensional visual sensory information and robot trajectory/path. This paper presents a dataset of successful needle insertion by da Vinci Research Kit into deformable objects based on which several deep-LfD models are built as a benchmark of models learning robot controller for the needle insertion task.