Self-Supervised Learning of Spatially Varying Process Parameter Models for Robotic Finishing Tasks

This paper presents a self-supervised learning approach for a robot to learn spatially varying process parameter models for contact-based finishing tasks. In many finishing tasks, a part has spatially varying stiffness. Some regions of the part enable the robot to efficiently execute the task. On the other hand, some other regions on the part may require the robot to move cautiously in order to prevent damage and ensure safety. Compared to the constant process parameter models, spatially varying process parameter models are more complex and challenging to learn. Our self-supervised learning approach consists of utilizing an initial parameter space exploration method, surrogate modeling, selection of region sequencing policy, and development of process parameter selection policy. We showed that by carefully selecting and optimizing learning components, this approach enables a robot to efficiently learn spatially varying process parameter models for a given contact-based finishing task. We demonstrated the effectiveness of our approach through computational simulations and physical experiments with a robotic sanding case study. Our work shows that the learning approach that has been optimized based on task characteristics significantly outperforms an unoptimized learning approach based on the overall task completion time.