This review article presents recent advancements in deep learning methodologies and applications for autonomous navigation. It analyzes state-of-the-art deep learning frameworks used in tasks like signal processing, attitude estimation, obstacle detection, scene perception, and path planning. The implementation and testing methodologies of these approaches are critically evaluated, highlighting their strengths, limitations, and areas for further development. The review emphasizes the interdisciplinary nature of autonomous navigation and addresses challenges posed by dynamic and complex environments, uncertainty, and obstacles. With a particular focus on mobile robots, self-driving cars, unmanned aerial vehicles, and space vehicles to underscore the importance of navigation in these domains. By synthesizing findings from multiple studies, the review aims to be a valuable resource for researchers and practitioners, contributing to the advancement of novel approaches. Key aspects covered include the classification of deep learning applications, recent advancements in methods, general applications in the field, innovations, challenges, and limitations associated with learning-based navigation systems. This review also explores current research trends and future directions in the field. This extensive overview, initiated in 2020, provides a valuable resource for researchers of all levels, from seasoned experts to newcomers. Its main purpose is to streamline the process of identifying, evaluating, and interpreting relevant research, ultimately contributing to the progress and development of autonomous navigation technologies.

Inertial Measurement Units (IMU) are commonly used in inertial attitude estimation from engineering to medical sciences. There may be disturbances and high dynamics in the environment of these applications. Also, their motion characteristics and patterns also may differ. Many conventional filters have been proposed to tackle the inertial attitude estimation problem based on IMU measurements. There is no generalization over motion and environmental characteristics in these filters. As a result, the presented conventional filters will face various motion characteristics and patterns, which will limit filter performance and need to optimize the filter parameters for each situation. In this paper, two end-to-end deep-learning models are proposed to solve the problem of real-time attitude estimation by using inertial sensor measurements, which are generalized to motion patterns, sampling rates, and environmental disturbances. The proposed models incorporate accelerometer and gyroscope readings as inputs, which are collected from a combination of five public datasets. The models consist of convolutional neural network (CNN) layers combined with Bi-Directional Long-Short Term Memory (LSTM) followed by a Fully Forward Neural Network (FFNN) to estimate the quaternion. To evaluate the validity and reliability, we have performed an extensive and comprehensive evaluation over five publicly available datasets, which consist of more than 120 hours and 200 kilometers of IMU measurements. The results show that the proposed method outperforms the state-of-the-art methods in terms of accuracy and robustness. Furthermore, it demonstrates that this model generalizes better than other methods over various motion characteristics and sensor sampling rates.