This research paper delves into the field of quadrotor dynamics, which are famous by their nonlinearity, under-actuation, and multivariable nature. Due to the critical need for precise modeling and control in this context we explore the capabilities of NARX (Nonlinear AutoRegressive with eXogenous inputs) Neural Networks (NN). These networks are employed for comprehensive and accurate modeling of quadrotor behaviors, take advantage of their ability to capture the hided dynamics. Our research encompasses a rigorous experimental setup, including the use of PRBS (Pseudo-random binary sequence) signals for excitation, to validate the efficacy of NARX-NN in predicting and controlling quadrotor dynamics. The results reveal exceptional accuracy, with fit percentages exceeding 99% on both estimation and validation data. Moreover, we identified the quadrotor dynamics using different NARX NN structures, including the NARX model with a sigmoid NN, NARX feedforward NN, and cascade NN. In summary, our study positions NARX-NN as a transformative tool for quadrotor applications, ranging from autonomous navigation to aerial robotics, thanks to their accurate and comprehensive modeling capabilities.

翻译：本研究深入探讨了四旋翼动力学领域，该领域以其非线性、欠驱动及多变量特性著称。鉴于在此背景下精确建模与控制的迫切需求，我们探索了NARX（非线性自回归外生输入）神经网络（NN）的能力。这些网络被用于全面且精确地建模四旋翼行为，充分利用其捕捉隐含动力学的优势。我们的研究涵盖了严谨的实验设置，包括使用PRBS（伪随机二进制序列）信号进行激励，以验证NARX-NN在四旋翼动力学预测与控制中的有效性。结果显示出了卓越的精度，在估计和验证数据上的拟合百分比均超过99%。此外，我们采用不同的NARX神经网络结构辨识了四旋翼动力学，包括具有S型神经元的NARX模型、NARX前馈神经网络以及级联神经网络。总之，凭借其精确且全面的建模能力，我们的研究将NARX-NN定位为四旋翼应用（从自主导航到空中机器人）的变革性工具。