Trajectory generation for robotic assembly operations using learning by demonstration

Abstract

The number of industrial robots and the type of industrial applications for which they are used, such as assembly operations, has increased considerably in the last few decades. However, programming continues to be a task which requires a high level of technical knowledge, and which is quite time consuming. This document proposes a methodology for the execution of parts assembly operations, using Learning by Demonstration (LbD) techniques, such as TP-GMM (Task Parametrized Gaussian Mixture Model). Task demonstrations were carried out by a person, and were visually acquired using a Kinect motion sensor. Later, these were used to train probabilistic models. Petri nets were used for the automatic generation of assembly plans, from those parts detected by the Kinect. Finally, operations were simulated in RobotStudio software, and later performed by an ABB IRB 140 robot. The results obtained demonstrate that it is possible to use programming techniques, which are different from traditional methodology such as LbD, to obtain satisfactory results in trajectories generation for simple assembly operations.