**1. Introduction**

Significant research is being conducted on the development of autonomy for underwater robotic vehicles, which are widely employed in many fields of application such as oceanographic, marine archeology [1], military organizations, and cable tracking and inspection [2].

© 2016 The Author(s). Licensee InTech. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and eproduction in any medium, provided the original work is properly cited. © 2018 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

The advent of underwater robotic vehicles has significantly reduced the dangers in deep sea exploration. Two kinds of robotic vehicles used in marine research are remotely operated vehicles (ROVs) and autonomous underwater vehicles (AUVs). The main difference between the two is that ROVs are connected to the ship by communication cables whereas AUVs operate independently from the ship. AUVs operate without an umbilical; therefore, AUVs are able to conduct a larger range of work area. In this research, AUV intelligent rescue system (AIRS) can minimize the loss of losing or unpredictable catastrophic failure of AUVs caused by emergency conditions.

In recent years, the micro-electro-mechanical system (MEMS) has almost become a vital technology for modern society because of its small volume, low power consumption, low cost, and ease of integration into systems or modification [3]. The MEMS technology creates entirely innovative kinds of products, such as gyroscope sensor in camera-shake detection systems [4], multi-axis inertial motion sensors for smartphone-based navigation [5], and rehabilitation systems based on inertial measurement units (IMUs) [6]. Additionally, the integration of global positioning system and inertial navigation system (GPS/INS) is usually employed to measure the position of AUVs [7, 8]. Unfortunately, small errors in the measurement of initial data are double integrated into larger errors progressively in attitude data, and such errors increase without bound. Error reduction calibration for initialization of INS is paramount for the systematic parameter, like scale factor, bias, and misalignment of the axes. As the situation depicted above, a calibration method is investigated and adopted in this research. There are many researches utilizing Kalman filter (KF) [9], complementary filter (CF) [10], adaptive Kalman filter (AKF) [11], or extended Kalman filter (EKF) [12] to fuse the gyroscope and accelerometer together, taking advantages of their individual strength. In this research, we use EKF to filter IMU outputs with a balance of noise canceling and adaptability simultaneously used in sensing attitude algorithm for AIRS.

The sensing algorithm with EKF for sensing the attitude of AUV, while employing the backpropagation network (BPN) to classify motion data that are formed in the AUV. Before classify the data, there is a problem. The problem is compounded by the fact that our system is computationally complex due to values estimation of AUV and causes the high dimension that is called "curse of dimensionality," which donates the drastic rise of computational complexity and the classification error. Therefore, the step of feature extraction (FE) is an important procedure in data classification [26]. Further, we feed the data into eural network to distinguish every AUV motion in each operating period. The neural network consists of multiple artificial neurons to receive inputs, and process them to obtain an output [13]. By repeating amendments to the model weights, the neural network makes central processing unit (CPU) more logical to calculate the nonlinear systems of AUV motion [14]. The neural network has been widely used because of a number of advantages, including estimating which variables are important in classification, detecting possible interactions between predictor variables, and constructing the prediction model with a high accuracy rate. Furthermore, the neural network has emerged as an important tool in classification which has been investigated in many different important applications [15]. In this study, neural network classifier (NNC) is used in the calculation of nonlinear systems AUV attitude data to detect a catastrophic failure of an AUV. Based on the multilayer feed-forward backpropagation algorithm as NNC, we proposed an effective activity recognition method using 10-DOF sensor module.

The concept of AIRS is inspired from the vehicle safety device: airbags have a flexible fabric bag that can protect and restrict the occupant from being thrown away during a crash accident [16, 17]. Wang and Dragcevic designed the airbags, which fire via a small pyrotechnic charge to increase motorcyclists' safety protections when riding a motorcycle [18, 19]. The airbags are applied to not only fall-protection devices but also automatic inflatable life buoy by the inflatable method of a gas cylinder [20–23]. We extended the airbag systems to load on the AUV with CO2 cylinder to implement the AIRS. In this research, we developed a malformed detection algorithm for AUV with both EKF and BPN, and the implementation of the airbag system loaded on AUV is underway.
