We propose an IMFAC method, which requires only the I/O data of the system to achieve precise control of the exoskeleton. Meanwhile, an ISO algorithm is proposed to search for the optimal initial conditions of the IMFAC method.

We propose a graph neural architecture search framework to build GNN model for disorder prediction. We use the GNNExplainer method to provide the explainability of the model, and the explainability results provide valuable insights for diagnosis and treatment.

We propose an innovative auto-learning search framework (ALSF) based on a weighted double Q-learning (WDQ-learning) algo-rithm. The development of this framework is tailored to opti-mize the construction of neural networks to achieve superior gesture recognition performance.

We propose a hierarchical exploration mechanism based on reinforcement learning to automatically generate high-performance CNN models, which can be used in detecting the muscle fatigue.

We propose a multi-feature temporal convolutional attention-based network (MFTCAN) to estimate the joint angles.
We develop a joint training mechanism that integrates MFTCAN with traditional statistical algorithms such as KNR and SVR.

We develop a robust SampEn-based algorithm to overcome the influence of the motion artifacts and irregular tonic spikes.
We design a double threshold adaptive detection frame-work with interlocking structure to detect the onsets and offsets of muscle activation intervals.

We propose the rapid refined composite multiscale sample entropy (R2CMSE) to extract sEMG signal features, and this algorithm can describe the change process of muscle fatigue.