隨著全球人口高齡化與行動障礙人數逐年增加，下肢外骨骼輔具逐漸成為提升行動能力的重要輔助技術。傳統的延遲輸出回饋控制（Delayed Output Feedback Control, DOFC）策略具備即時、平順的輔助能力，但其延遲參數設定通常需手動調整，無法即時適應使用者個體差異與動作狀態變化。為克服此限制，本研究提出結合深度學習之自適應DOFC策略，以實現下肢外骨骼系統之智慧化、個人化控制。本研究使用慣性測量單元（MPU6050）及AK60-6馬達編碼器，擷取步態運動訊號，並以表面肌電訊號（sEMG）分析使用者的肌肉負擔。透過卷積神經網路（CNN）與雙向長短期記憶網路（BiLSTM）之深度學習模型，於不同步行速度（4 km/h與5 km/h）下自動學習與預測最佳DOFC延遲參數。實驗結果顯示，所提出的自適應控制策略在不同步態速度下，能有效學習最佳延遲時間。以4 km/h速度進行測試時，延遲參數最佳值為0.9秒；而5 km/h速度下則為0.7秒。此外，本研究建立之CNN-BiLSTM模型表現出高精度之扭矩預測能力，證明本系統具備良好準確性。本研究自製EMG電路，採用AD8421放大器與濾波處理，擷取並即時分析肌電訊號，評估外骨骼對肌肉負擔的輔助成效。

As the global population ages and the number of individuals with mobility impairments continues to rise, lower-limb exoskeleton assistive devices are increasingly becoming a crucial assistive technology for enhancing mobility. Traditional Delayed Output Feedback Control (DOFC) strategies offer real-time and smooth assistive capabilities; however, their delay parameters typically require manual adjustment and cannot adapt in real-time to individual user differences or changes in movement states. To overcome this limitation, this study proposes an adaptive DOFC strategy combined with deep learning to achieve intelligent and personalized control of lower limb exoskeleton systems.
This study uses an inertial measurement unit (MPU6050) and an AK60-6 motor encoder to capture gait movement signals and analyze the user's muscle load using surface electromyography (sEMG) signals. Through a deep learning model combining a convolutional neural network (CNN) and a bidirectional long short-term memory network (BiLSTM), the system automatically learns and predicts the optimal DOFC delay parameters at different walking speeds (4 km/h and 5 km/h). Experimental results show that the proposed adaptive control strategy effectively learns the optimal delay time at different gait speeds. When tested at 4 km/h, the optimal delay parameter value was 0.9 seconds; at 5 km/h, it was 0.7 seconds. Additionally, the CNN-BiLSTM model developed in this study demonstrated high-precision torque prediction capabilities, proving the system's excellent accuracy.
This study developed an EMG circuit using AD8421 amplifiers and filter processing to capture and analyze electromyographic signals in real time to evaluate the effectiveness of exoskeletons in assisting muscle load.

[1] A. J. Young and D. P. Ferris, "State of the art and future directions for lower limb robotic exoskeletons," IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 25, no. 2, pp. 171-182, 2016.
[2] S. K. Banala, S. H. Kim, S. K. Agrawal, and J. P. Scholz, "Robot assisted gait training with active leg exoskeleton (ALEX)," IEEE transactions on neural systems and rehabilitation engineering, vol. 17, no. 1, pp. 2-8, 2008.
[3] H. Kawamoto, S. Kanbe, and Y. Sankai, "Power assist method for HAL-3 estimating operator's intention based on motion information," in The 12th IEEE International Workshop on Robot and Human Interactive Communication, 2003. Proceedings. ROMAN 2003., 2003: IEEE, pp. 67-72.
[4] A. Zoss, H. Kazerooni, and A. Chu, "On the mechanical design of the Berkeley Lower Extremity Exoskeleton (BLEEX)," in 2005 IEEE/RSJ international conference on intelligent robots and systems, 2005: IEEE, pp. 3465-3472.
[5] Y. Tu, A. Zhu, J. Song, X. Zhang, and G. Cao, "Design and experimental evaluation of a lower-limb exoskeleton for assisting workers with motorized tuning of squat heights," IEEE transactions on neural systems and rehabilitation engineering, vol. 30, pp. 184-193, 2022.
[6] C. Caulcrick, W. Huo, W. Hoult, and R. Vaidyanathan, "Human joint torque modelling with MMG and EMG during lower limb human-exoskeleton interaction," IEEE Robotics and Automation Letters, vol. 6, no. 4, pp. 7185-7192, 2021.
[7] W. Huo, S. Mohammed, J. C. Moreno, and Y. Amirat, "Lower limb wearable robots for assistance and rehabilitation: A state of the art," IEEE systems Journal, vol. 10, no. 3, pp. 1068-1081, 2014.
[8] Y. Long and Y. Peng, "Extended state observer-based nonlinear terminal sliding mode control with feedforward compensation for lower extremity exoskeleton," IEEE Access, vol. 10, pp. 8643-8652, 2021.
[9] Z. Ding et al., "The real time gait phase detection based on long short-term memory," in 2018 IEEE Third International Conference on Data Science in Cyberspace (DSC), 2018: IEEE, pp. 33-38.
[10] I. P. Pappas, M. R. Popovic, T. Keller, V. Dietz, and M. Morari, "A reliable gait phase detection system," IEEE Transactions on neural systems and rehabilitation engineering, vol. 9, no. 2, pp. 113-125, 2001.
[11] A. Narayan, F. A. Reyes, M. Ren, and Y. Haoyong, "Real-time hierarchical classification of time series data for locomotion mode detection," IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 4, pp. 1749-1760, 2021.
[12] W. Choi, W. Yang, J. Na, J. Park, G. Lee, and W. Nam, "Unsupervised gait phase estimation with domain-adversarial neural network and adaptive window," IEEE Journal of Biomedical and Health Informatics, vol. 26, no. 7, pp. 3373-3384, 2021.
[13] J. Yin, T. Xue, and T. Zhang, "Real-Time Gait Trajectory Prediction Based on Convolutional Neural Network with Multi-Head Attention," in 2022 27th International Conference on Automation and Computing (ICAC), 2022: IEEE, pp. 1-6.
[14] S. Bala and D. Joshi, "An attention-based deep CNN-BiLSTM model for forecasting of fatigue-induced surface electromyography signals during isotonic contractions," Authorea Preprints, 2022.
[15] D. Zhao et al., "Upper limb human-exoskeleton system motion state classification based on semg: application of CNN-BiLSTM-attention model," Scientific Reports, vol. 15, no. 1, p. 18969, 2025.
[16] B. Lim et al., "Delayed output feedback control for gait assistance with a robotic hip exoskeleton," IEEE Transactions on Robotics, vol. 35, no. 4, pp. 1055-1062, 2019.
[17] P. Konrad, "The abc of emg," A practical introduction to kinesiological electromyography, vol. 1, no. 2005, pp. 30-5, 2005.
[18] C. J. De Luca, "Surface electromyography: Detection and recording," DelSys Incorporated, vol. 10, no. 2, pp. 1-10, 2002.
[19] A. Rainoldi, G. Melchiorri, and I. Caruso, "A method for positioning electrodes during surface EMG recordings in lower limb muscles," Journal of neuroscience methods, vol. 134, no. 1, pp. 37-43, 2004.
[20] H. Yali and W. Xingsong, "Biomechanics study of human lower limb walking: Implication for design of power-assisted robot," in 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2010: IEEE, pp. 3398-3403.
[21] 林育銘, "智能行動外骨骼輔具之電路系統整合設計及電量估測," 碩士論文, 電機工程學系, 國立成功大學, 台南市, 2024. [Online]. Available: https://hdl.handle.net/11296/w88g84
[22] J. L. Elman, "Finding structure in time," Cognitive science, vol. 14, no. 2, pp. 179-211, 1990.
[23] S. Hochreiter and J. Schmidhuber, "Long short-term memory," Neural computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[24] J. Chung, C. Gulcehre, K. Cho, and Y. Bengio, "Empirical evaluation of gated recurrent neural networks on sequence modeling," arXiv preprint arXiv:1412.3555, 2014.
[25] W. J. Magrath, C. S. Qiu, P. J. Hanwright, S. H. Tuffaha, and N. Khavanin, "A systematic review of muscle synergies during a walking gait to define optimal donor-recipient pairings for lower extremity functional reconstruction," Plastic and Reconstructive Surgery–Global Open, vol. 10, no. 8, p. e4438, 2022.
[26] T. Instruments, "Single-supply, 2nd-order, Sallen-Key low-pass filter circuit," Analog Engineer’s Circuit, 2021.
[27] T. Instruments, "Single-supply, 2nd-order, Sallen-Key high-pass filter circuit," Analog Engineer’s Circuit, 2021.
[28] A. Dehghani, O. Sarbishei, T. Glatard, and E. Shihab, "A quantitative comparison of overlapping and non-overlapping sliding windows for human activity recognition using inertial sensors," Sensors, vol. 19, no. 22, p. 5026, 2019.
[29] K. Chen, L. Kurgan, and J. Ruan, "Optimization of the sliding window size for protein structure prediction," in 2006 IEEE Symposium on Computational Intelligence and Bioinformatics and Computational Biology, 2006: IEEE, pp. 1-7.
[30] G. M. Ólafsdóttir, "Exploring the potential of a model-based approach to detect muscular dysfunction using surface EMG," 2023.