本研究以低成本慣性感測器為主體，針對無人水面載具在速度估測時常見的量測雜訊、零偏漂移與積分誤差問題，建立一套多感測器融合之速度估測與誤差抑制方法。系統整合兩組慣性量測單元（Inertial Measurement Unit, IMU）之加速度量測，並搭配由推進器佔空比(Duty Cycle)輸入建立對應相對船速的目標預測速度，形成多來源速度資訊架構，目標是在不依賴昂貴設備的前提下，穩定提供導航所需的速度與位置資訊。
在訊號處理流程中，首先針對加速度資料進行偏移修正與低通濾波，抑制水槽環境中的高頻震動與感測器隨機雜訊，接著透過非線性死區處理，濾除零附近的小幅度抖動，以減少積分運算時的誤差累積。對於單純加速度積分易產生速度漂移的問題，本研究利用零速更新（Zero-Velocity Update, ZUPT）及條件式速度校正機制，於推進指令低於啟動門檻或偵測到靜止狀態時，將速度強制歸零，並在速度與參考值偏差過大且伴隨加速度峰值時進行平滑修正。
經前處理後，可分別從兩組IMU與目標預測速度獲得多組速度估測結果。本文以自適應權重融合策略整合這些速度來源：先以各路徑速度與整體平均速度的偏離程度評估可信度，偏差越大的來源，其權重越低，再將加權後的速度輸出作為卡爾曼濾波之量測輸入，進一步平滑短時間波動並得到連續的速度估測序列。最後結合慣性航向資訊，進行平面航跡的航向推算（Dead Reckoning, DR），以評估整體導航性能。
實驗於國立成功大學拖航水槽進行，分別以平均船速0.38m/s、0.71m/s與1.00m/s進行測試，比較不同方法之速度與距離估測誤差。結果顯示，相較於直接對原始加速度積分，加入偏移修正、低通濾波、非線性死區與ZUPT後，可明顯抑制速度發散現象；在三種航速條件下，多感測器融合加卡爾曼濾波所得到的速度誤差維持在相對較低的水準，且未出現明顯漂移。雖然在部分條件下，融合結果的數值誤差不必然最低，但在感測器量測差異較大或單一路徑出現異常時，融合方法可維持較平穩的輸出與容錯能力。整體而言，本研究所提出之多感測器速度估測架構在拖航水槽實驗中具有實作可行性，可作為低成本無人水面載具導航系統中速度與位置估測的基礎。

This study focuses on low-cost inertial sensors and develops a multi-sensor fusion method for velocity estimation and error suppression to address common problems encountered in velocity estimation of unmanned surface vehicles (USVs), including measurement noise, bias drift, and integration errors. The proposed system integrates acceleration measurements from two inertial measurement units (IMUs) and incorporates a target predicted velocity established from the propeller duty cycle to construct a multi-source velocity information architecture. The objective is to provide stable velocity and position information required for navigation without relying on expensive sensing equipment.
In the signal processing procedure, acceleration data are first subjected to bias correction and low-pass filtering to suppress high-frequency vibrations and random sensor noise in the towing tank environment. Nonlinear dead-zone processing is then applied to remove small-amplitude fluctuations around zero, thereby reducing error accumulation during integration. To address the velocity drift caused by direct acceleration integration, zero-velocity update (ZUPT) and a conditional velocity correction mechanism are employed. When the propulsion command is below the activation threshold or a stationary state is detected, the velocity is reset to zero. In addition, when the deviation between the estimated velocity and the reference value becomes excessive and is accompanied by an acceleration peak, a smoothing correction is performed.
After preprocessing, multiple velocity estimates are obtained from the two IMUs and the target predicted velocity. An adaptive weighting fusion strategy is employed to integrate these velocity sources. The reliability of each velocity source is evaluated based on its deviation from the overall mean velocity, with larger deviations resulting in lower weights. The weighted velocity output is then used as the measurement input of a Kalman filter to further smooth short-term fluctuations and obtain a continuous velocity estimation sequence. Finally, inertial heading information is combined with the estimated velocity to perform planar trajectory estimation using dead reckoning (DR), thereby evaluating the overall navigation performance.
Experiments were conducted in the towing tank at National Cheng Kung University under three average vessel speeds of 0.38 m/s, 0.71 m/s, and 1.00 m/s. The velocity and distance estimation errors of different methods were compared under these operating conditions. The results show that, compared with direct integration of raw acceleration data, the incorporation of bias correction, low-pass filtering, nonlinear dead-zone processing, and ZUPT can effectively suppress velocity divergence. Under all three velocity conditions, the velocity errors obtained using multi-sensor fusion combined with the Kalman filter remained relatively low, without significant drift. Although the fusion results did not necessarily produce the lowest numerical error under all conditions, the proposed method maintained smoother outputs and improved fault tolerance when significant discrepancies occurred between sensor measurements or when an individual sensing path exhibited abnormal behavior. Overall, the proposed multi-sensor velocity estimation architecture demonstrated practical feasibility in towing tank experiments and can serve as a foundation for velocity and position estimation in low-cost navigation systems for unmanned surface vehicles.

[1] Bar-Shalom, Y., Li, X. R., & Kirubarajan, T. L. (2001). Estimation with applications to tracking and navigation. Wiley.
[2] Brown, R. G., & Hwang, P. Y. C. (2012). Introduction to random signals and applied Kalman filtering (4th ed.). John Wiley & Sons.
[3] Butterworth, S. (1930). On the theory of filter amplifiers. Experimental Wireless & the Wireless Engineer, 7, 536–541.
[4] Chui, C. K., & Chen, G. (2017). Kalman filtering: With real-time applications (5th ed.). Springer.
[5] Çoçoli, A., & Badia, F. (2023). Comparative analysis of Kalman, Mahony and Madgwick filters for attitude estimation in low-cost IMUs. Proceedings of the 2023 International Seminar on Intelligent Technology and Its Applications (ISITIA), 160–165.
[6] Dong, Y., Wu, N., Qi, J., Chen, X., & Hua, C. (2021). Predictive course control and guidance of autonomous unmanned sailboat based on efficient sampled Gaussian process. Journal of Marine Science and Engineering, 9(12), 1420.
[7] Fossen, T. I. (2011). Handbook of marine craft hydrodynamics and motion control. Wiley.
[8] Groves, P. D. (2013). Principles of GNSS, inertial, and multisensor integrated navigation systems (2nd ed.). Artech House.
[9] Karimi, A., Jalilvand, A., & Shakouri, A. (2013). Performance comparison of EKF and UKF for INS/DVL integration in underwater navigation. Proceedings of the IEEE International Conference on Robotics and Mechatronics (ICRoM), 1127–1132.
[10] Kartidjo, M., Nugroho, S. A., & Wibowo, W. (2015). Design and analysis of CentrUMS-ITB unmanned semi-submersible vehicle. Proceedings of the 2015 10th Asian Control Conference (ASCC). IEEE.
[11] Kuipers, J. B. (1999). Quaternions and rotation sequences: A primer with applications to orbits, aerospace, and virtual reality. Princeton University Press.
[12] Liggins, M. E., II, Hall, D. L., & Llinas, J. (Eds.). (2009). Handbook of multisensor data fusion: Theory and practice (2nd ed.). CRC Press.
[13] Liu, C.-P., Chen, C.-Y., Chen, Y.-J., & Lin, H.-Y. (2024). Low-cost underwater integrated navigation system using MEMS IMU, DVL, and USBL sensors. Journal of Marine Science and Engineering, 12(2), 200.
[14] Mu, M., Ji, L., & Zhang, Y. (2021). Adaptive fault-tolerant integrated navigation algorithm based on variational Bayesian and multi-sensor fusion for AUV. Applied Ocean Research, 110, 102598.
[15] Savage, P. G. (2007). Strapdown analytics II, Part 1: Introduction to strapdown inertial navigation systems. Strapdown Associates.
[16] Shaukat, N., Ali, A., Iqbal, M. J., Moinuddin, M., & Otero, P. (2021). Multi-sensor fusion for underwater vehicle localization by augmentation of RBF neural network and error-state Kalman filter. Sensors, 21(3).
[17] Titterton, D. H., & Weston, J. L. (2004). Strapdown inertial navigation technology (2nd ed.). IET (Institution of Engineering & Technology).
[18] 李怡萱（2012）。三維數位資訊輔助INS/GPS無縫式導航演算法之效益分析〔碩士論文，國立成功大學〕。
[19] 陳鳴遠、胡竹生（2013）。使用多個狀態限制卡爾曼濾波器與三焦張量幾何之視覺輔助慣性測程器〔碩士論文，國立交通大學〕。
[20] 曾庭鴻（2002）。以GPS鑑別加速儀的速度誤差模式〔碩士論文，國立成功大學〕。
[21] 黃敬傑、彭政中（2018）。基IMU之迭代加權Levenberg–Marquardt校正法與AHRS演算法之開發與驗證〔碩士論文，國立成功大學〕。