VISION-BASED ROBOTIC PICK-AND-HANDOVER MANEUVERS USING VARIABLE IMPEDANCE CONTROL

Abstract

The adoption of advanced intelligent manipulator systems to carry out pick-and-handover tasks has seen a rise in both the manufacturing and healthcare sectors, thanks to their impressive precision, adaptability, and operational efficiency. Execution of these tasks demands the synergistic functioning of various modules, encompassing the sensor system used for gathering environmental data, the control algorithm used for manipulator, and the trajectory planning algorithm. The primary objective of this thesis focuses on building a framework, which integrates these modules. A significant merit of this framework is its inherent capacity for easier upgrades. Due to its modular structure, it allows for the modification or replacement of individual components without causing any disruptions to the overall system. Initially, a vision-based impedance control method is employed with a 7-degree-of-freedom (7-DOF) Franka Emika (FE) Panda robotic manipulator to accomplish pick-and-handover tasks, featuring human-like fruit grasping capabilities, which ensures a basic framework with different modules is built. Subsequently, two low cost vision modules are established for both two-dimensional (2D) and three-dimensional (3D) object recognition, localization, and anthropomorphic manipulation, utilizing the You Only Look Once version five (YOLOv5) system. The salient attribute of this segment revolves around attaining a commendable level of accuracy using low-cost cameras. The innovative modular-based framework also allows for a smooth transition between different types of camera modules, such as shifting from a standard camera module to a depth camera module or a laser module. To facilitate the end-effector in picking up objects with varying characteristics, a novel Difference-based Dynamic Movement Primitives (DMPs) algorithm is utilized for trajectory planning module to generate human-like trajectories. Finally, a variable impedance controller is designed for control module to strike an optimal balance between precision, safety, and efficiency during the object pick-and-handover task.