| 摘要: | 本研究針對市售輪椅動力輔助系統成本高、體積大、拆裝複雜與缺乏模組化彈性等問題，設計並完成一套小型化、輕量化且具可拆裝能力之動力輔助模組，並將控制架構寫入嵌入式單晶片(STM32)中，實現即時運算之控制實驗系統。
動力輔助模組採雙摩擦輪設計，能分別對應單側輪胎輸出正反向扭矩，具備傳遞效率高與安裝彈性佳的特性。控制策略方面，本研究整合輪速量測技術、驅動力估測模型與智慧輔助控制機制，並針對微控制器之運算與記憶體配置進行最佳化，以確保多通道訊號處理與即時且高效之晶片性能需求。本研究使用之輪速量測技術在低速條件下依然具備良好解析度，最低可精確量測之輪速達0.175 rad/s，能滿足手動輪椅於日常推行之低速需求。由於系統設計採用摩擦輪與輪胎間的齒比放大效應，轉速訊號在傳輸至感測端後得以有效提升解析度與穩定性，進一步改善了低速區間下的量測精度。透過使用者實際推動輪椅所獲得之輪速曲線進行頻域分析可知，推動手輪的主要運動能量集中於0–10 Hz的低頻範圍，由於嵌入式控制器僅能進行序列控制，輸出多通道PWM訊號時可能產生微小時間差；然而，頻域結果顯示該時間差與使用者實際運動頻率相差一萬倍，對整體運動行為影響可忽略不計。
本研究於單晶片中，使用驅動力估測模型獲得使用者扭矩，取代成本昂貴之扭矩感知器，控制策略方面，將透過驅動力估測模型與智慧輔助控制邏輯，協助使用者安全抵達目的地。透過實驗驗證確認本系統於不同推進情境皆具備良好之即時性與可靠性。實驗結果顯示，在直線行駛、變換車道及S型轉彎等多種操作情境下，本系統能即時依據驅動力估測模型準確推算使用者施加於手輪之扭矩，並透過強化學習架構動態調整輔助策略，使輔助力輸出更能符合使用者實際需求。與未啟動輔助之情境比較，實驗數據顯示使用者施加於手輪的推進扭矩平均降低約21 %，顯著減輕上肢負荷並提升整體推進效率與操作舒適性。此外，系統能即時響應不同操作路徑的動態需求，確保使用者於各種環境下均能穩定獲得輔助輸出，進一步提升操作安全性與任務完成效率。主觀評估結果中，受測者普遍認為輔助力輸出平順且連續，不會產生突兀的感受，使得操作體驗更趨接近自然推行，同時大幅減少長時間操作所帶來的疲勞感。
綜合而言，本研究提出之模組在結構設計、控制策略與實驗驗證方面均展現實際應用可行性，能有效降低上肢負荷並提升日常推行效能，具備推廣至不同使用者與多樣環境下應用的潛力。;This study addresses the challenges of commercial power-assist wheelchair systems, including high cost, bulky size, complicated assembly, and lack of modularity. To overcome these limitations, we developed a compact, lightweight, and detachable power-assist module, integrating the control architecture into an embedded STM32 microcontroller to achieve real-time control and execution.
The module employs a dual-friction-wheel design capable of independently applying bidirectional torque to each rear wheel, offering high transmission efficiency and flexible installation. The control strategy integrates wheel speed sensing, a driving force estimation model, and an intelligent assist control mechanism. Computational and memory optimization on the microcontroller ensures efficient multi-channel signal processing and real-time performance. The wheel speed sensing method maintains high resolution even at low speeds, capable of accurately detecting speeds as low as 0.175 rad/s, which meets the requirements of typical manual wheelchair propulsion. Additionally, the friction-driven gear ratio amplifies the wheel speed signal before measurement, significantly improving resolution and stability in low-speed conditions.
Frequency-domain analysis of real-world wheelchair propulsion reveals that most motion energy is concentrated below 10 Hz. Although multi-channel PWM generation on the embedded controller introduces minimal temporal offsets, these are negligible compared to the user’s movement frequency. Instead of expensive torque sensors, this system estimates user-applied torque using a model-based approach implemented in the microcontroller. The controller dynamically adapts assistance strategies based on real-time torque estimation to help users safely reach their destinations.
Experimental validation in various scenarios—including straight driving, lane changes, and S-curve maneuvers—demonstrates the system’s accuracy, responsiveness, and robustness. Compared to the unassisted condition, the average propulsion torque applied by users was reduced by approximately 21%, effectively lowering upper limb strain and improving overall propulsion efficiency and comfort. The system also delivers stable assistance across different trajectories, enhancing both safety and task performance. Subjective user evaluations indicated that the assistive torque felt smooth and natural, significantly reducing fatigue during prolonged use.
In conclusion, the proposed module exhibits practical applicability in its mechanical design, control strategy, and experimental performance. It effectively reduces physical demand while improving daily mobility, offering strong potential for broader adoption across diverse users and environments. |
