6D Visual Servoing for Industrial Manipulators applied to Human-Robot Interaction Scenarios

Abstract

This thesis presents a novel vision-based control system for industrial manipulators. The primary contribution is an uncalibrated Visual Servoing (VS) based on 6D orthogonal features. These features have good linearization and decoupling properties, leading to a full-rank image Jacobian which allows avoiding classical VS problems such as image space singularities and local minima. By integrating the VS system with the environment and robot model constraints in real world applications using a prioritized multi-constraints control framework, we demonstrate that it can be easily and safely integrated into a variety of robotic systems involving human-robot interaction.