本論文使用資料驅動的方式搭配上 LSTM 模型，求得近似的控制系統轉移函數，並對模型進行修正、行為分析與極點零點之比對。
為了驗證此方法的可靠性，我們先於虛擬的系統進行實驗，再將此論文的方法套用在實體馬達系統上，虛擬控制系統的驗證方式如下：首先建立數個階數不同的受控體 (Plant) 與對應之 PID 控制器兩個元件的簡單負回授系統，以多種不同的輸入訊號作為測試，並收集巨量的測試資料，以之完成 LSTM 模型訓練後進行此模型的系統頻率響應之轉換，並經過正規化後與上述系統的轉移函數進行極點與零點的比較，證明此一方法的可行性後，隨後在使用伺服馬達之實體動態系統中同樣以透過資料驅動的方式求得系統轉移函數，以利後續控制器的調整以及控制訊號的設計。
本論文另涵蓋使用 ROS2 環境之數位孿生平台，測試並模擬移位機器人及人體機器人模型在 Unity 物理引擎驅動下的移動情形並進行受力分析，作為安全性評估的依據，未來將整合此學習模式於 Unity 中移位機器人的馬達控制，以達到完整模擬實體場域中的移位機器人之控制情形。

This work employs a data-driven approach combined with an LSTM model to approximate the transfer function of a control system. The model is then refined, analyzed in terms of system behavior, and compared based on its poles and zeros.
To verify the reliability of this method, experiments are first conducted in a virtual environment before applying the approach to a physical motor system. The validation process in the virtual control system is as follows: several simple negative feedback systems are constructed, each consisting of a plant with different system orders and a corresponding PID controller. Various input signals are used for testing, and a large amount of data is collected. After training the LSTM model with this dataset, the system’s frequency response is obtained and normalized. The resulting transfer function is then compared with that of the original system by analyzing poles and zeros to demonstrate the feasibility of the proposed method. Once validated, the same data-driven approach is applied to a physical dynamic system using a servo motor to estimate its transfer function, facilitating subsequent controller tuning and control signal design.
Additionally, this work includes the development of a digital twin platform using the ROS2 environment to test and simulate the movement of a transfer-assist robot and a human-like robot model under Unity’s physics engine. A force analysis is conducted as the basis for safety evaluation. In the future, this learning-based approach will be integrated into the motor control of the transfer robot in Unity, enabling full simulation of its behavior as it would occur in a real-world environment.

[1] Khadija El Hamidi, Mostafa Mjahed, Abdeljalil El Kari, and Hassan Ayad. Adap-tive control using neural networks and approximate models for nonlinear dynamic systems. Modelling and Simulation in Engineering, 2020(1):8642915, 2020.
[2] Emirhan Inanc, Yigit Gurses, Abdullah Habboush, Yildiray Yildiz, and Anu-radha M. Annaswamy. Neural network adaptive control with long short-term memory, 2023.
[3] R. Pintelon and J. Schoukens. System Identification: A Frequency Domain Ap-proach. Wiley, 2012.
[4] L. Ljung. System Identification: Theory for the User. Prentice Hall, 1999.
[5] T. S¨oderstr¨om and P. Stoica. System Identification. Prentice-Hall, 1989.
[6] Tiham´er ´Ad´am, Samad Dadvandipour, and J´ozsef Fut´as. Influence of discretiza-tion method on the digital control system performance. Acta Montanistica Slovaca, 8(4):197–200, 2003.
[7] Sepp Hochreiter and J¨urgen Schmidhuber. Long short-term memory. Neural computation, 9(8):1735–1780, 1997.
[8] Febina Christudas et al. System identification using long short term memory for dynamic control systems. ROMJIST, 2020. LSTM model for laboratory-scale nonlinear dynamic systems, applied to controller design.
[9] Kaicheng Niu, Mi Zhou, Chaouki T. Abdallah, and Mohammad Hayajneh. Deep transfer learning for system identification using long short-term memory neural networks, 2022.
[10] Yu Wang. A new concept using lstm neural networks for dynamic system identifica-tion. In 2017 American Control Conference (ACC), 2017. Proposes convex-based LSTM for faster dynamic system identification.
[11] Morgan Quigley, Josh Faust, Tully Foote, Jeremy Leibs, et al. Ros: an open-source robot operating system. In ICRA Workshop on Open Source Software, 2009.
[12] Steven Macenski, Tully Foote, Brian Gerkey, Chris Lalancette, and William Woodall. Robot operating system 2: Design, architecture, and uses in the wild. Science robotics, 7(66):eabm6074, 2022.
[13] Tobias Kronauer, Joshwa Pohlmann, Maximilian Matth´e, Till Smejkal, and Ger-hard Fettweis. Latency analysis of ros2 multi-node systems. In 2021 IEEE inter-national conference on multisensor fusion and integration for intelligent systems (MFI), pages 1–7. IEEE, 2021.
[14] Jan Staschulat, Ralph Lange, and Dakshina Narahari Dasari. Budget-based real-time executor for micro-ros. arXiv preprint arXiv:2105.05590, 2021.
[15] Michael Grieves. Digital twin: Manufacturing excellence through virtual factor replication. 03 2015.
[16] John Smith, Alice Wang, and Robert Lee. Robotics manipulation and control using unity-based virtual environments. Journal of Robotics and Automation, 36(4):123–135, 2020.
[17] Hyun Lee, Min Kim, and Sarah Chen. Digital twin for smart factory: Real-time monitoring and predictive maintenance using unity. In Proceedings of the IEEE International Conference on Industrial Informatics, pages 567–572, 2021.
[18] Wei Chen, Ling Zhang, and Michael Brown. Sensor fusion for mobile robot navi-gation: A unity simulation approach. International Journal of Intelligent Robotics Research, 45(7):987–998, 2019.
[19] Cheng-Chieh Liu. A ros2-based digital twin system for robot control and interac-tion in unity. Master’s thesis, National Cheng Kung University, Tainan, Taiwan,R.O.C., 2024. Available from National Digital Library of Theses and Dissertations in Taiwan, https://hdl.handle.net/11296/fn6255.
[20] CAN Specification. Bosch. Robert Bosch GmbH, Postfach, 50:15, 1991.