The Iterative Learning Gain That Optimizes Real-Time Torque Tracking for Ankle Exoskeletons in Human Walking Under Gait Variations

Lower-limb exoskeletons often use torque control to manipulate energy flow and ensure human safety. The accuracy of the applied torque greatly affects how well the motion is assisted and therefore improving it is always of interest. Feed-forward iterative learning, which is similar to predictive stride-wise integral control, has proven an effective compensation to feedback control for torque tracking in exoskeletons with complicated dynamics during human walking. Although the intention of iterative learning was initially to benefit average tracking performance over multiple strides, we found that, after proper gain tuning, it can also help improve real-time torque tracking. We used theoretical analysis to predict an optimal iterative-learning gain as the inverse of the passive actuator stiffness. Walking experiments resulted in an optimum gain equal to 0.99 ± 0.38 times the predicted value, confirming our hypothesis. The results of this study provide guidance for the design of torque controllers in robotic legged locomotion systems and will help improve the performance of robots that assist gait.