Reproducing arm movements based on Pose Estimation with robot programming by demonstration

Teaching robot movements has always been considered a complex topic in which there is much interest, since the slightest change in the robot's programming can generate a high downtime that can last for months. In this work, we carried out a study of human movements to implement a new method of Robot Programming by Demonstration (RPbD) using neural networks. Current methods require specialists with high mathematical and logical knowledge to teach robot movements. Using a famous pose estimation algorithm called OpenPose and a 3D lifting method we obtain the estimated pose of the person arm in a simulated 3D space. Then, we use various classification tools to translate it to the robot. The results show that it is feasible to make robot programming more accessible using pose estimation.