.png){kind=icon}
.png){kind=icon}
Return
Hill-Derived Muscle Dynamics-Driven Multi-Task Framework With Gates Control Network for Lower-Limb Exoskeleton Joint Angle Estimation in Class-Imbalanced Scenarios
.png){kind=icon}
PRE
AI
H
X
F
Z
E
Z
P
DOI：10.1109/tim.2026.3716463.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate decoding of continuous lower-limb movements from surface electromyography (sEMG) is crucial to improving exoskeleton control compliance. Most studies focus on enhancing cross-subject and cross-scenario generalization. However, mainstream data-driven methods rely heavily on large labeled datasets and overlook class-imbalanced scenarios caused by the high cost of data acquisition in real-world applications. To address this challenge, we propose a multi-task framework for estimating lower-limb exoskeleton multijoint angles, aiming to mitigate the negative impact of class imbalance in the data. Specifically, a Hill-derived muscle dynamics feature extraction algorithm is proposed to enhance physiological interpretability by modeling muscle components. Furthermore, a multi-task learning model with a gate control network is proposed to adaptively fuse task-specific and task-shared knowledge, reducing majority-class bias and alleviating class imbalance. Ten participants were recruited to perform experiments involving six walking tasks, which included four walking speeds and two slopes. Compared with state-of-the-art models, the proposed framework achieved lower root-mean-square error (RMSE) and higher correlation coefficient (CC) in both class-balanced and class-imbalanced conditions. Specifically, in class-imbalanced conditions, the RMSE of the knee and hip joints ranged from 2.32° to 2.94° and from 1.94° to 2.84°, respectively. This study mitigates the class imbalance resulting from high data acquisition costs, thereby enhancing the applicability of wearable lower-limb exoskeletons in complex real-world scenarios.
Keywords:
Class-imbalanced scenarios
gate control network
hill muscle model (HMM)
joint angle estimation
multi-task learning
Journal
IF:
5.9
Papers:
1.9W
Citations:
5.8W
Organization
Cited Papers