上肢外骨骼施力估測Upper-Limb Exoskeleton Force Estimation

於國科會「具適應性之主被動式上肢外骨骼機器人系統開發及其於復健之應用」計畫擔任兩年研究助理。上肢外骨骼若要協助中風或肌力退化患者復健，關鍵在於「讀懂使用者想出多少力」——唯有正確估測人體的施力意圖，外骨骼才能適時提供主動輔助或被動阻抗，避免過度代償。我負責的核心子題，就是以肌電（EMG）訊號為基礎，建立人體施力的估測與姿態識別模型。A two-year research-assistant role on an NSTC-funded adaptive active-passive upper-limb exoskeleton project. For an upper-limb exoskeleton to aid stroke or muscle-weakness rehabilitation, the key is reading how much effort the user intends to exert — only with an accurate estimate of human force intent can the exoskeleton supply active assistance or passive resistance at the right moment without over-compensating. My core sub-task was to build force-estimation and posture-recognition models from electromyography (EMG) signals.

• 訊號擷取：以表面肌電（sEMG）感測器量測上肢主要肌群的活化訊號。Acquisition: surface EMG (sEMG) sensors capture activation from the main upper-limb muscle groups.
• 前處理：於 MATLAB 進行帶通濾波、整流、移動 RMS 平滑與最大自主收縮（MVC）正規化，抑制雜訊與個體差異。Pre-processing: band-pass filtering, rectification, moving-RMS smoothing, and MVC normalization in MATLAB to suppress noise and inter-subject variation.
• 施力估測：建立 EMG 活化量與實際關節施力之對應模型，輸出連續的施力估測值。Force estimation: a model mapping EMG activation to actual joint effort, outputting a continuous force estimate.
• 姿態識別：依肌群活化模式辨識動作意圖（如屈曲／伸展），供外骨骼控制器決策輔助時機。Posture recognition: classifies movement intent (e.g., flexion/extension) from activation patterns to inform the exoskeleton controller's assist timing.

兩年期間我負責訊號處理流程與估測模型的開發與驗證，並協助實驗資料的蒐集與分析。這段經歷讓我紮實地建立生醫訊號處理、特徵工程與人機協作控制的基礎，也是我日後投入 AI 與訊號相關研究的起點。Over two years I owned the signal-processing pipeline and the development and validation of the estimation models, and supported experimental data collection and analysis. The work gave me a solid foundation in biomedical signal processing, feature engineering, and human-robot cooperative control — and set the stage for my later AI and signal research.