Dyadic behavior in co-manipulation :from humans to robots

Abstract

To both decrease the physical toll on a human worker, and increase a robot’s environment perception, a human-robot dyad may be used to co-manipulate a shared object. From the premise that humans are efficient working together, this work’s approach is to investigate human-human dyads co-manipulating an object. The co-manipulation is evaluated from motion capture data, surface electromyography (EMG) sensors, and custom contact sensors for qualitative performance analysis. A human-human dyadic co-manipulation experiment is designed in which every human is instructed to behave as a leader, as a follower or neither, acting as naturally as possible. The experiment data analysis revealed that humans modulate their arm mechanical impedance depending on their role during the co-manipulation. In order to emulate the human behavior during a co-manipulation task, an admittance controller with varying stiffness is presented. The desired stiffness is continuously varied based on a scalar and smooth function that assigns a degree of leadership to the robot. Furthermore, the controller is analyzed through simulations, its stability is analyzed by Lyapunov. The resulting object trajectories greatly resemble the patterns seen in the human-human dyad experiment.