VISION BASED LEARNING OF GRIPPER TRAJECTORIES FOR A ROBOT ARM.

Abstract

As opposed to the usual approach of programming robot arms by giving a complete description of the trajectory (e.g. direct programming) this work is based on the idea of Programming by Demonstration (PbD). Only few intermediate positions, which approximate the trajectory, are given and recorded by a stereo-camera-system. The idea is to extract the path of the manipulator from the images by geometrical connecting this positions to form a smooth trajectory. This is done by tracking the appearance of the manipulator in the sequence of stereo images. The manipulator position in previous images and the manipulator gripper appearance are used for locating it in the next images. Based on this sequence of positions a trajectory-structure is formed, which allows to execute a vision based movement. The trajectory is general in the sense that the starting position and the orientation can be specified variable. The approach can be used in automotive industry for vision based learning of trajectori...