Deep-Learning-Based Neural Network Training for State Estimation Enhancement: Application to Attitude Estimation

Abstract

Achieving precise state estimation is needed for the unmanned aerial vehicle to perform a successful flight with a high degree of stability. Nonetheless, obtaining accurate state estimation is considered challenging due to the inaccuracies associated with the measurements of the onboard commercial-offthe-shelf (COTS) Inertial Measurement Unit (IMU). The immense vibration of the vehicle's rotors makes these measurements suffer from issues like; large drifts, biases and immense unpredictable noise sequences. These issues cannot be significantly tackled using classical estimators and an accurate sensor fusion technique needs to be developed. In this paper, a deep learning framework is developed to enhance the performance of the state estimator. A deep neural network (DNN) is trained using a deep-learning-based technique to identify the associated measurement noise models and filter them out. Dropout technique is adopted for training the DNN to avoid overfitting and reduce the complexity of nets computations. Compared to the classical estimation results, the proposed deep learning technique demonstrates capabilities in identifying the measurement's noise characteristics. As an example, an enhancement in estimating the attitude states at near hover is proved using this approach. Furthermore, an actual hover flight was performed to validate the proposed estimation enhancement method.

The New Trend of State Estimation: From Model-Driven to Hybrid-Driven Methods

“…The last one is to obtain the estimation state directly [ 115 , 116 , 117 , 118 ]. For example, Gao used measurement data as input and the reference state as output data to train the LSTM network [ 115 ]; Bai used the NARX (Nonlinear autoregressive with external input) network to learn the complex relationship between the measurement, state, and filter gain in nonlinear systems [ 116 ].…”

A Stacked LSTM-Based Approach for Reducing Semantic Pose Estimation Error

“…This functionality cannot be realized when using conventional neural networks. DNNs and LSTMs have been exhibiting state-of-the-art performance in a multitude of various applications, including computer vision [12], [13], [14] and robotics [15], [16], [17].…”

“…The effect of unpredictable nonuniform noise as well as external environmental conditions is also inevitable [35]. To enhance the accuracy of localization, the solutions found in the literature can be classified into: (1) controlling the environment under investigation [36], (2) sensor data fusion [37], [38], (3) improving measurement covariane estimation [39], [30], [35], [40], or (4) correcting measurement errors, which can be further classified into classical [41], [42], [43], [44] and learning approaches [16], [45], [34], [46], [17], [47].…”

“…In [16], a deep learning based approach is employed to improve the altitude estimation of a flying robotic vehicle. Moreover, in [45], the accuracy of the odometry of a wheeled cart, calculated using its dynamic equations, was improved using an SNN.…”

EKF-based self-attitude estimation with DNN learning landscape information

“…In [10], IMU-based odometry by end-toend learning has been proposed. In [11], a deep neural network identifies the measurement noise characteristics of IMU. In [12], a neural network estimates angular rates from sequential images.…”