In a complex underwater environment, finding a viable, collision-free path for an autonomous underwater vehicle (AUV) is a challenging task. The purpose of this paper is to establish a safe, real-time, and robust method of collision avoidance that improves the autonomy of AUVs. We propose a method based on active sonar, which utilizes a deep reinforcement learning algorithm to learn the processed sonar information to navigate the AUV in an uncertain environment. We compare the performance of double deep Q-network algorithms with that of a genetic algorithm and deep learning. We propose a line-of-sight guidance method to mitigate abrupt changes in the yaw direction and smooth the heading changes when the AUV switches trajectory. The different experimental results show that the double deep Q-network algorithms ensure excellent collision avoidance performance. The effectiveness of the algorithm proposed in this paper was verified in three environments: random static, mixed static, and complex dynamic. The results show that the proposed algorithm has significant advantages over other algorithms in terms of success rate, collision avoidance performance, and generalization ability. The double deep Q-network algorithm proposed in this paper is superior to the genetic algorithm and deep learning in terms of the running time, total path, performance in avoiding collisions with moving obstacles, and planning time for each step. After the algorithm is trained in a simulated environment, it can still perform online learning according to the information of the environment after deployment and adjust the weight of the network in real-time. These results demonstrate that the proposed approach has significant potential for practical applications.

Visual odometry (VO) is one of essential techniques in mobile robots field, an accurate VO system is of great significance for mobile robot simultaneous localization and mapping. As for traditional monocular VO systems, they work by presuming the monocular scale is 1 (scale=1), or relying on ground truth (GT) to estimate scale. As a result, the traditional monocular VO systems estimate the pose state with big drift or cannot work on the image sequence without GT. Although some classical monocular VO systems have been proposed, however, they still have imperfect performance or even unable to work in some extreme scene conditions, such as scene is monotony without obvious texture information or camera large-scale displacement motion. As for learning-based VO system, it is realized by training deep neural networks in supervised or self-supervised manner to end-to-end estimate the pose state, however, the accuracy of pose estimation entirely depends on the ability of networks. Although the ability of networks can be improved by increasing the number of training datasets and optimizing the network structure, it is inevitable to encounter problems such as insufficient generalization ability and insufficient accuracy on rotational pose estimation. In this paper, a monocular VO system named DL_Hybrid is proposed, which takes full advantage of DL networks used in image processing and geometric localization theory based on hybrid pose estimation methods. The DL_Hybrid VO system can estimate a six-DoF pose one-frame-by-one-frame and recover camera trajectory, and it can extract accurate key points from per-frame even in extreme scene condition, and it has good performance even in the extreme moving condition, such as camera rotation-only action or static action, also it can work well in the condition of camera large-scale displacement motion. The real scale is also accurately estimated without depending on GT, and the pose estimation method is designed based on hybrid 2d-2d&3d-2d localization theory to make the DL_Hyrid VO system to estimate translational and rotational information with accuracy and robustness. Experiment results show that the proposed DL_Hybrid VO system has better performance than traditional and learning-based VO systems.

Target state estimation is a key technology for unmanned underwater vehicles to achieve target tracking, collision avoiding, formation control, and other tasks. Compared with other measurement methods, underwater measurement has lower reliability due to the uncertainty of sonar detection. In this case, the performance of target state estimation depends heavily on the target motion model. However, the dynamics of unmanned underwater vehicles are very complex and nonlinear. Although many state estimation methods for nonlinear systems have been proposed, the complex dynamics of unmanned underwater vehicle and uncertainty in sonar detection remain challenges for underwater target state estimation problems. This paper proposes a gated recurrent units-based particle filter to improve the performance of the target state estimator for unmanned underwater vehicles. A deep neural network framework based on gated recurrent units is used to establish the mapping between the previous measurements and the current target states. This neural network learns the dynamics model of unmanned underwater vehicles and recognizes the measurement noise. The proposed filter samples from previous measurements of the target unmanned underwater vehicle, and the fully trained deep neural network predicts the current states of the sampled particles. The proposed method solves the low accuracy and instability caused by motion modeling errors and system nonlinearity. Simulation results show that the proposed gated recurrent units-based particle filter for unmanned underwater vehicle state estimation has better accuracy and stability than the traditional state estimation methods. All simulations are done with the noise which is Gaussian noise pass through a nonlinear transformation.

At present, the tracking accuracy of underwater passive target tracking is often limited due to models that are overly simple, with low complexity, poor universality, and an inability to learn. In this paper, a cubature Kalman filter (CKF) algorithm based on a gated recurrent unit (GRU) network is proposed. The filter innovation, prediction error, and filter gain obtained from the CKF are used as the input to the GRU network, and the filter error value is used as the output to train the network. End-to-end online learning is carried out using the designed fully connected network, and the current state of the target is predicted. In this paper, a deep neural network based on the GRU architecture is used to convert the tracking prediction problem into a time series prediction problem in the field of artificial intelligence, and its strong fitting ability is used to resolve the uncertainty of the target motion. Simulation results show that an unmanned underwater vehicle (UUV) state estimation method based on the GRU filter proposed in this paper offers better accuracy and stability than the traditional state estimation method.