Selecting the Best Series Stiffness and Iterative Learning Gain for Exoskeleton Torque Control

Series elastic actuators have been widely used in humanrobot interactive systems due to their high controllability and improved human experience. One such system is the powered ankle exoskeleton. In [1], the torque control of a tethered ankle exoskeleton with a coil spring (Fig. 1) was investigated, which showed that the combination of model-free, integration-free feedback control and iterative learning resulted in the best tracking performance. During these torque control experiments, interesting interactions were observed between the exoskeleton hardware, the high level controller that determines desired torques and the low level controller that tracks torques. In this study, we focus on the interactions between three stiffness or stiffness-like values in ankle exoskeleton torque control: the desired stiffness Kdes defined as the slope of desired torque τdes and ankle angle θa relationship, the passive stiffness Kt defined as the relationship between the torque τ and angle θa while the motor position is fixed, and the iterative learning gain Kl. Hypotheses based on symbolic analysis are made and tested through walking experiments. The results of this project are expected to answer these questions: What is the optimal hardware compliance for a fixed control objective? What is the optimal iterative learning gain for a fixed combination of system compliance and control objective? 2 Hypotheses