Few-Shot Learning of a Force-Based Industrial Cleaning Process using an Instrumented Tool

Abstract

Force-based processes like cleaning, grinding, and polishing are essential in industrial and domestic applications but challenging to automate with robots, particularly in high-mix/low-volume scenarios. These tasks require precise replication of tool poses, forces, and velocities, making the teach-in complex and tedious. This work presents a system for robotic surface cleaning that uses an instrumented tool and a base plate for mechanical alignment to capture high-quality human demonstrations and the interaction between the tool and the surface. The proposed few-shot learning framework is location invariant and independent of the specific robot platform, significantly reducing the need for extensive demonstrations. The simulation and experimental results show that the system generalizes to objects with similar geometric features utilizing few human demonstrations, providing an efficient solution for industrial applications.