Human-robot collaboration for object manipulation in manufacturing

Abstract

In this thesis, we explore the task of transporting bulky objects of different shape and rigidity, which is currently done by multiple human workers manually. By deploying robot(s) with advanced control strategies, we present three manufacturing scenarios that may benefit from using both a human and robot collaboratively in the task.

First, a multi-robot system is implemented to stably grasp and transport a large load through frictional contacts. A human operator provides motion guidance by directly manipulating the load. Through contact force sensing, the multi-arm/multi-robot system infers the human intent for motion while maintaining the contact force closure condition. The contact forces at the end-effectors need to be carefully managed to avoid possible loss of contact. This is particularly challenging during motion, where the motion-induced inertial force adds a disturbance to the static stable grasp condition. We propose a feedforward force compensation scheme using support vector regression to estimate the inertial force. This enables the robots to coordinate their motion to maintain contacts, avoid slippage, and follow human hand-guided motion.

We then integrate a sensor-driven dual-arm mobile robot system to semi-automate a composite sheet layup task. By monitoring the force feedback as well as visual and verbal information, the proposed robot system can follow the lead of a human operator to transport a large deformable sheet to a designated location and help the operator to conform the sheet onto a mandrel. The robot's two end-effectors that grasp the deformable sheet can automatically adjust their spatial velocities to maintain the desired geometry and surface compliance of the sheet and follow the lead of the human operator while avoiding self-collision and singularity.

Finally, we conduct a feasibility study of robotic fixtureless assembly. Human-guided path planning for the robot with a load is critical to ensure collision-free motion in a tight space. Machine vision is used for panel localization and pick-up, and visual servoing for placement using fiducial markers. To avoid damage to the panel, compliance force control augments the vision guidance for human supervising control.