GenJAPNet: A Generalizable Joint Angle Prediction Network with Non-Redundant Muscle Synergy Features for Lower-Limb Exoskeletons
Hairong Zhang, Yu Bai, Kou Ziming, Wu Juan, Pengjie Qin, Fei GAO, Wenze Shang, yue Teng, Dingkui Tian, Xinyu Wu
AI summary
Problem
Accurate lower-limb joint angle prediction from sEMG signals struggles with inter-subject variability and differing locomotion speeds, limiting exoskeleton adaptability. Existing feature extraction and modeling methods often suffer from redundancy and poor generalization to unseen users or conditions.
Approach
The method extracts non-redundant muscle synergy features from sEMG using NMF and UMAP, then processes them through a pre-trained ResNet and TCN-BiLSTM architecture with a meta-learner. Few-shot fine-tuning allows the model to rapidly adapt to new walking speeds or subjects with minimal data.
Key results
- NMF-UMAP features outperform traditional time-domain features with 15–52% lower RMSE
- Robust cross-speed and cross-subject joint angle prediction accuracy
- Rapid adaptation to new speeds or subjects using only ~30% of data
- Validated feasibility through physical exoskeleton walking experiments
Why it matters
Provides a robust, data-efficient control framework that enhances the adaptability and clinical deployment of lower-limb exoskeletons across diverse users and walking conditions.
Abstract
Lower-limb exoskeleton robots play a significant role in both rehabilitation and assisted walking, where ac- curate prediction of lower-limb joint angles is crucial for achieving natural gait. However, due to inter-subject variability and differences across locomotion modes, achieving cross- task generalization in joint angle prediction remains a major challenge. This work proposes a novel framework for multi- joint angle prediction in the lower-limb, which includes a non- redundant muscle synergy feature extraction algorithm and a Generalizable Joint Angle Prediction Network (GenJAPNet) across speeds and subjects. The feature extraction algorithm employs Non-negative Matrix Factorization (NMF) to extract activation coefficient matrix from Surface Electromyography (sEMG) signals, followed by further dimensionality reduc- tion using Uniform Manifold Approximation and Projection (UMAP) to obtain more discriminative and non-redundant features. GenJAPNet leverages pre-trained shared features and few-shot fine-tuning to rapidly adapt to new task. Through feature extraction algorithm comparison experiments, cross- speed and cross-subject experiments, and exoskeleton-assisted walking physical experiments, the effectiveness and generaliz- ability of this method are validated, demonstrating its potential for enhancing the performance of lower-limb exoskeleton reha- bilitation and assistive applications.