Deep Learning-Driven Analysis of a Six-Bar Mechanism for Personalized Gait Rehabilitation

Journal of Computing and Information Science in Engineering, Journal Year: 2024, Volume and Issue: 25(1)

Published: Oct. 14, 2024

Abstract Recent advances in robotics and artificial intelligence have highlighted the potential for integration of computational enhancing functionality adaptability robotic systems, particularly rehabilitation. Designing exoskeletons lower limb rehabilitation post-stroke patients requires frequent adjustments to accommodate individual differences leg anatomy. This complex engineering challenge necessitates a deep understanding human physiology, robotics, optimization develop adaptive systems also swiftly quantify required implement them each patient. The conventional approaches, which mostly rely on heuristics manual tuning, often struggle achieve optimal results. paper presents novel method that integrates genetic algorithm with learning approach generate gait trajectory ankle joint from six-bar linkage mechanism fixed dimensions. Later, using same approach, inverse kinematics solution this is devised whereby, set link dimensions obtained given an customization. We simulated kinematic behavior within defined mechanical constraints utilized generated data training feedforward neural network long short-term memory models. proposed model, when trained, can produce accurate lengths desired trajectories sagittal plane vice versa, further validates our solution. Moreover, evaluate efficiency models, we conducted extensive error-based, comparative, sensitivity analysis different performance indices. results highlight deep-learning-driven design robots.

Language: Английский

---

Obstacle Feature Information-Based Motion Decision-Making Method for Obstacle-Crossing Motions in Lower Limb Exoskeleton Robots

Biomimetics, Journal Year: 2025, Volume and Issue: 10(5), P. 311 - 311

Published: May 12, 2025

To overcome the problem of insufficient adaptability to motion environment lower limb exoskeleton robots, this paper introduces computer vision technology into control robots and studies an obstacle-crossing-motion method based on detecting obstacle feature information. Considering information different obstacles distance between a trajectory planning direct point matching was used generate offline adjusted gait libraries obstacle-crossing libraries. A robot decision-making algorithm is proposed by combining constraints constraints, enabling it select appropriate trajectories in library. The validated at three distances with four obstacles. experimental results show that can from library detected safely complete motions.

Language: Английский

---

EEG-TCNTransformer: A Temporal Convolutional Transformer for Motor Imagery Brain–Computer Interfaces

Signals, Journal Year: 2024, Volume and Issue: 5(3), P. 605 - 632

Published: Sept. 23, 2024

In brain–computer interface motor imagery (BCI-MI) systems, convolutional neural networks (CNNs) have traditionally dominated as the deep learning method of choice, demonstrating significant advancements in state-of-the-art studies. Recently, Transformer models with attention mechanisms emerged a sophisticated technique, enhancing capture long-term dependencies and intricate feature relationships BCI-MI. This research investigates performance EEG-TCNet EEG-Conformer models, which are trained validated using various hyperparameters bandpass filters during preprocessing to assess improvements model accuracy. Additionally, this study introduces EEG-TCNTransformer, novel that integrates architecture series self-attention blocks employing multi-head structure. EEG-TCNTransformer achieves an accuracy 83.41% without application filtering.

Language: Английский

---

EEG-TCNTransformer: A Temporal Convolutional Transformer for Motor Imagery Brain-Computer Interfaces

Published: Aug. 9, 2024

In Brain-Computer Interface Motor Imagery (BCI-MI) systems, Convolutional Neural Networks (CNNs) have traditionally dominated as the deep learning method of choice, demonstrating significant advancements in state-of-the-art studies. Recently, Transformer models with attention mechanisms emerged a sophisticated technique, enhancing capture long-term dependencies and intricate feature relationships BCI-MI. This research investigates performance EEG-TCNet EEG-Conformer models, which are trained validated using various hyperparameters bandpass filters during preprocessing to assess improvements model accuracy. Additionally, this study introduces EEG-TCNTransformer, novel that integrates convolutional architecture series self-attention blocks employing multi-head structure. The EEG-TCNTransformer achieves an accuracy 82.97% without application filtering. source code is available on GitHub.

Language: Английский

---

Method of automatic biomedical signals interpretation for safety supervision and optimisation of the exoskeleton-aided physiotherapy of lower extremity

Published: June 27, 2024

Language: Английский

---