Vision Based Robotic Traversal of Textureless Smooth Surfaces

This study investigates visually localizing and servoing a serial robot’s tooling assembly with respect to a smooth, contoured workpiece surface through the dynamic extraction and leveraging of characteristic surface invariants. A novel formulation for 6 degree-of-freedom (DoF) textureless visual servoing is proposed which significantly extends an existing 3-DoF scheme designed for planar surfaces. A new family of observed feature sets, and their associated interaction matrices are proposed. Curvature conditions on the workpiece shape are derived under which local asymptotic stability for 6-DoF is guaranteed.

The approach naturally, and synergistically, combines the geometric tools of computer aided design, path generation, and computer numerical control (CAD/CAM/CNC) theory, with the manipulator control tools of visual servoing. Design focuses on ease of use by accepting task descriptions in the form of time-parameterized, surface-constrained curves which are compatible with the output of existing path generation systems. Thus, this work is a contribution to the growing collection of integrated-process manufacturing techniques streamlining and extending the traditional CAD/CAM/CNC pipeline. Full surface-relative 6-DoF pose control is achieved with respect to a workpiece which need not be precisely located, or even stationary, in the workspace. Preliminary considerations are also made for a non-rigid workpiece.