Multi-Sensor Fusion-Based Mirror Adaptive Assist-as-Needed Control Strategy of a Soft Exoskeleton for Upper Limb Rehabilitation
Ning Li,Yang Yang,Gang Li,Tie Yang,Yihan Wang,Wenyuan Chen,Peng Yu,Xiujuan Xue,Chuang Zhang,Wenxue Wang,Ning Xi,Lianqing Liu
DOI: https://doi.org/10.1109/tase.2022.3225727
IF: 6.636
2022-01-01
IEEE Transactions on Automation Science and Engineering
Abstract:Assist-as-needed (AAN) assistance can promote active voluntary participation in rehabilitation and motor function recovery of post-stroke patients. However, different patients have personalized damaged regions and recovery states, causing difficulties to obtain adaptive and customized assistance in robot-assisted rehabilitation. This paper presents a mirror Adaptive Assist-As-Needed (AAAN) scheme, including two modules of Multi-Sensors Fused Estimation (MSFE) and Online Incremental Mirror Adaptation (OIMA), to encourage the subjects to actively participate in rehabilitation. Specifically, the first MSFE module can obtain the needed assistance based on the functional capability of the post-stroke patients via the data fusion of biological and motional signals using Kalman Filter. The second OIMA module fine-tunes the control torques estimated by MSFE to adapt the muscle fatigue and stiffness varieties of the affected limb based on the motion and physiological reference of the mirror healthy limb. The results demonstrate that the AAAN strategy can realize the transparent mode for healthy subjects and promote post-stroke patients to rehabilitate the affected limb with active participation using EMG signals 90.5% similar to those of the mirror healthy limb. The proposed method can be expected to greatly enhance power assistance and rehabilitation outcome of post-stroke patients using exoskeletons by provoking active participation. Note to Practitioners—For robotic rehabilitation, it is crucial to provide suitable assistances that can maximize the participation of post-stroke patients, which can promote the recovery outcome of therapies. The main purpose of this work is to achieve the adaptive assist-as-needed control strategy for upper limb rehabilitation tasks in two steps. Firstly, the elbow joint torques of a post-stroke patient are estimated by data fusion of motion and electromyography (EMG) signals using Kalman Filter, which can make up for the shortcomings of the individual signals, such as poor reliability and low sensitivity. Secondly, the joint motion and EMG signals of the mirror healthy limb are used as the reference to calculate the adaptive needed assistance to rehabilitate the affected limb. The preliminary experiments with healthy and post-stroke subjects demonstrate that this approach can obtain stable motion with the natural physiological states of subjects and enhance active voluntary participation in rehabilitation. In the future study, it will be investigated how to accelerate the adaptation of new patients based on the knowledge of the learned individuals using machine learning methods, such as lifelong learning and incremental learning.
automation & control systems