Training by Demonstration for Welding Robots by Optical Trajectory Tracking

The paper shows the application of an optical measuring system to a trajectory tracking problem with the aim of training a welding robot in a collaborative man-machine welding station. The optical system allows a fast and reliable training of the welding tasks imparted by the human co-worker. An algorithm developed for this application filters inaccurate and noisy data, re-orders points and computes a trajectory suitable for a welding robot. This can shorten time required for programming. The industrial relevance of this application is twofold: on one side it increases the safety of human workers by leaving the robot to perform the risky tasks, namely the welding, on the other side it guarantees higher repeatability than manual execution.