Abstract
In this paper, we study a new method for robot programming: view-based teaching/playback. The motivation of its development is to achieve more robustness against changes of task conditions than conventional teaching/playbackwithout losing its general versatility. For proof of concept, the method was implemented and tested on a virtual environment. The method is composed of two parts: teaching phase and playback phase. In the teaching phase, a human operator commands a robot to achieve a manipulation task. All themovements of the robot are recorded. All the images of the teaching scenes are also recorded. Then, a mapping from the recorded images to the movements is obtained as an artificial neural network. In the playback phase, the motion of the robot is determined by the output of the neural network using scene images. We applied this view-based teaching/playback to pick-and-place and pushing in a virtual environment. Human demonstrated manipulation was successfully reproduced by a virtual robot hand with our proposed method. Moreover, manipulation of the object fromsome initial positions that are not identical to those in the demonstrations was also successfully achieved with our method.
