Preference-Based Assistance Map Learning With Robust Adaptive Oscillators

Recently, lower-limb exoskeletons have demonstrated the ability to enhance human mobility by reducing biological efforts with human-in-the-loop (HIL) optimization. However, this technology is confined to the laboratory, and it is difficult to generalize to daily applications where gaits are more complex and professional equipment is not accessible. To solve this issue, firstly, we present a robust adaptive oscillator (RAO) to synchronize the human-robot movement and extract gait features. Then, we use the Gaussian process regression (GPR) to map the subjects' preferred assistance parameters to gait features. Experiments show that the RAO has a faster convergence rate compared with the traditional adaptive oscillators. Meanwhile, the learning efficiency of the proposed method shows superiority compared with the HIL optimization. The effectiveness of the proposed method is validated by a hip exoskeleton at a speed of 5 km/h with 7 participants. Three muscles which include rectus femoris, tibialis anterior, and medial gastrocnemius are investigated in three conditions: user-preferred assistance (ASS), zero torque (ZT), and normal walking (NW). The results show that all muscles achieve an activity reduction in ASS mode compared with ZT or NW. Meanwhile, there is a statistically significant difference on medial gastrocnemius in ASS mode with respect to both ZT and NW (-15.63 ± 6.51% and -8.73 ± 6.40%, respectively).