SurfMan: Generating Smooth End-Effector Trajectories on 3D Object Surfaces for Human-Demonstrated Pattern Sequence

At RA-L and IROS 2022 International Conference on Intelligent Robots and Systems, 2022

Abstract

Specifying robot tasks for low-volume manufacturing scenarios is an open problem. The state-of-the-art robotic systems enable the application of smooth 2D paths to a 3D surface but assume that the product engineer provides these paths. We extend this approach with a novel tool-path specification method, which produces smooth paths from noisy demonstrations. The user demonstrates only short patterns and selects a base path relative to an object in front of the robot, along which these patterns should be applied. The representation based on polynomials allows controlling the grade of the smoothness of the resulting tool path. We generate parametrized robot trajectories to meet the use-case-specific constraints and adhere to the robot's kinodynamic limits. We propose a set of measures to evaluate the quality of the generated curves and corresponding trajectories with respect to executability by a robot. The evaluation in simulation and real-robot experiments showed that the robot could reach up to 15.9% higher constant speed on tool paths generated by our system compared to unprocessed paths.