Trajectory generation for adhesive dispensing robots by modeling of material behavior

Abstract

It is difficult for robots to manipulate flexible objects, and adhesive dispensing is one such task. In this task, the adhesive material is pulled by a dispensing robot, which is problematic to predict. In this paper, we propose an analysis-based and a learning-based model to predict the behavior of the adhesive material, and a method to explore the robot trajectory. While analysis-based models consider physical behavior and require less training data, they are limited to specific physical behaviors. Learning-based models, on the other hand, can model many physical behaviors, but require a lot of training data. Finally, we use the predictions of these models to perform experiments and evaluate the differences between the target adhesive trajectory and the actual application results. • We proposed a method to predict adhesive behavior by two different models. • The analysis-based simple physical model was modeled with a small amount of data. • Learning-based linear time-series models were data-driven and thus more versatile. • Automation of trajectory generation with height variation presents challenges.