The challenge in contact-rich assembly tasks lies in the low tolerances of the tightly matching parts, as using stiff position controllers would lead to an excessive rise in contact forces in case of minor differences in part location.

Secondly, this project aims to increase the flexibility in automating such tasks, e.g., to quickly and cost-effectively adapt to different product variants. The objective of this work is to demonstrate and validate the developed Learning-from-Demonstration framework and the HARTU Application Manager addressing these challenges.

These solutions enable the robotic system to learn from a user’s kinesthetic teaching of motion and forces needed to perform an assembly step, to generalise this knowledge in terms of part locations, and to handle uncertainties in the part-joining phase by adapting its compliance.

Responsible partner: DFKI