To address the challenges posed by high-intensity noise and the structural characteristics of large obstacle targets in underwater sonar imaging, as well as the stringent requirements for lightweight deployment and high inference efficiency of perception algorithms in real-time underwater obstacle avoidance tasks, a semantic segmentation algorithm for sonar images with low computational cost and short inference time is proposed. The method aims to resolve the trade-off between the computational complexity of perception algorithms and the real-time response requirements in obstacle avoidance applications.

Based on an encoder-decoder network architecture, lightweight convolution operations were introduced to significantly reduce computational complexity. In addition, a large-kernel separable attention mechanism was incorporated into the skip connections to enhance feature fusion for obstacle avoidance scenarios. A dataset of 6936 sonar images collected and manually annotated from real environments was used for training and comparative experiments. Furthermore, the obstacle avoidance strategy based on the proposed perception algorithm was validated on the Gazebo simulation platform.

The improved algorithm specifically enhances the segmentation accuracy of large targets. Compared with the baseline model, the FLOP and the number of parameters are reduced by 69.2% and 83%, respectively. At the same time, the inference time is shortened by 22.6%, while perception accuracy improves by 10.8%. In addition, simulation experiments verify the effectiveness of the perception algorithm during the obstacle avoidance process, demonstrating that it fully satisfies the requirements of real-time perception tasks in underwater obstacle avoidance scenarios based on forward-looking sonar.

The proposed sonar-image-based perception algorithm can effectively meet the obstacle avoidance requirements of unmanned underwater vehicles in onboard operating scenarios and shows promising potential for engineering applications.

To address the challenge of simultaneously maintaining formation integrity and enabling flexible obstacle avoidance for multi-unmanned underwater vehicle (multi-UUV) formations in complex underwater environments, this paper proposes a global path planning method that supports adaptive formation reshaping.

The proposed method is built upon an affine transformation framework that maps the cooperative path planning problem of the multi-UUV system into a two-dimensional affine parameter space. First, a front-end path search is conducted using an improved rapidly-exploring random tree* (RRT*) algorithm. By integrating fast exploration and iterative optimization phases, a weighted k-dimensional (KD) tree, a hybrid sampling mechanism, and adaptive tuning of sampling parameters, this algorithm efficiently generates an initial sequence of affine states. Subsequently, a B-spline-based back-end optimizer employs a gradient descent method to minimize a comprehensive objective function that accounts for trajectory smoothness, UUV kinematic feasibility, environmental collision safety, and the cost associated with adaptive formation scaling. The optimization process yields a continuous and smooth trajectory of affine parameters that satisfies multiple constraints.

Lake experiments demonstrate that the proposed planning method can generate safe and feasible formation paths. It successfully guided the multi-UUV formation through a simulated narrow obstacle region, while the actual velocities and accelerations of the UUVs remained within the predefined feasibility constraints.

The proposed global planning method, based on affine transformation, effectively generates safe and feasible paths for multi-UUV formations navigating complex obstacle environments by enabling adaptive formation reshaping. This method significantly enhances the autonomy and environmental adaptability of marine unmanned vehicles, and holds great value for advancing the development and practical application of marine unmanned systems technology.

To address the high-performance trajectory tracking control problem for the unmanned underwater vehicle (UUV) under external disturbances, this paper proposes a non-singular integral terminal sliding mode control (NITSMC) scheme based on a nonlinear disturbance observer (NDOB).

First, the dynamic equation of UUV with lumped disturbances is established, and an NDOB is designed to estimate the lumped disturbances. On this basis, an NITSMC law is designed to effectively compensate for the disturbances, achieving fast and precise tracking of the desired trajectory. Finally, the effectiveness and superiority of the designed NDOB-NITSMC control scheme are validated through comparative simulations against the integral sliding mode control (ISMC) and PID control methods. Additionally, the effectiveness and practicality of NDOB-NITSMC are confirmed on the BlueROV2 UUV platform through water tank experiments.

The results show that, in high-fidelity simulations, the NDOB-NITSMC scheme ensures that the position and orientation errors of BlueROV2 converge to a small neighborhood around zero within 8 seconds, with root mean squared errors (RMSEs) of 0.31, 0.22, 0.29 centimeters, and 0.23 degrees, respectively. The convergence speed and control accuracy surpass those of ISMC and PID control methods. In water tank experiments, the NDOB-NITSMC scheme ensures that the position and orientation errors of the UUV converge to a small neighborhood around zero within 15 seconds, with RMSEs of 2.31, 2.56, 2.17 centimeters, and 1.11 degrees, respectively. Additionally, the control inputs are smooth, ensuring its suitability for practical engineering applications.

The proposed NDOB-NITSMC scheme effectively addresses the high-performance trajectory tracking control problem for UUVs under external disturbances and demonstrates promising prospects for engineering applications.

To improve the ability of autonomous underwater vehicle (AUV) formations to perform tasks in complex obstacle scenarios, a distributed three-dimensional affine formation shape maneuver control method is proposed for multi-AUV formations.

Based on the affine transformation theory and stress matrix, AUV formation shape can achieve rotation, scaling, shear, coplanarity, collineation, or their combination. Meanwhile, the yaw consistency of the multi-AUV formation is achieved based on the Laplacian matrix. Furthermore, a non-singular integral terminal sliding mode controller is employed to ensure that the multi-AUV formation can track the desired trajectory fast and with high accuracy.

Through high-fidelity simulation experiments, the proposed AUV formation maneuver control method can drive the multi-AUV formation to realize the high-precision maneuvering transformation of a three-dimensional formation with high maneuverability and strong robustness.

The proposed method enables the multi-AUV formation to avoid complex underwater obstacles by flexible maneuvering formation, which improves the safety of the AUV formation and its ability to perform tasks.

Inspired by remora, a robotic remora that can adhere to diverse hosts and travel with them over long distances with low energy consumption due to its unique adhesion ability, addresses the issue of low endurance in the robotic fish.

A prototype with decoupled three-dimensional motion, high mobility, and responsive vertical motion has been developed. Moreover, a wire-driven propulsion mechanism and a pectoral fin ascending and descending mechanism have been adopted to achieve high mobility and decoupled motion. Additionally, a central pattern generator (CPG) is adopted to realize high biomimetic swimming postures and smooth switching between motion modes. An active disturbance rejection controller (ADRC) is developed to achieve robust, fast and precise heading control under model uncertainty and the environment disturbance. The swimming, diving and heading control experiments are conducted.

The results exhibits that the designed robotic remora owns the high mobility and responsive decoupled motion capabilities, with the maximum swimming speed of 0.17 m/s and max ascent and descent speed of 0.095 and 0.099 5 m/s respectively. The designed ADRC heading controller can achieve fast and precise control, which is with better performance than the PID controller.

This study, which presents the design of the robotic remora and verifies the decoupled motion and the ADRC heading controller by experiments, lays the foundation for the auto adhesion of the robotic remora.

Considering that existing research on autonomous underwater vehicle (AUV) obstacle avoidance mainly focuses on low-speed obstacle avoidance for small and medium-sized AUVs and overly simplifies the diverse constraints within and outside the system, a real-time three-dimensional high-speed obstacle avoidance method for large AUVs is proposed.

The method integrates perception, planning, and control modules, enabling large-scale, high-speed, underactuated AUVs to navigate safely and efficiently through the unknown and unstructured ocean floor. First, a robocentric dual-resolution seafloor map is constructed to balance perception accuracy with computational efficiency. Next, a dynamic perception framework incorporating filters and feature extraction and matching is designed to achieve the motion prediction of unknown moving obstacles. Subsequently, global risk-aware path searching and local spatial-temporal trajectory optimization are introduced to generate an aggressive trajectory that satisfies multiple constraints. Finally, a spherical-coordinate feedback controller is employed for trajectory tracking.

In high-fidelity experiments involving long-range seabed traversal, a 13.96-meter-long AUV flexibly avoids dynamic and static obstacles while adhering to the constraints, maintaining a predefined speed of 6.0 m/s.

The proposed approach enables the large-scale high-speed AUV to navigate agilely and avoid obstacles safely under limited field of view and multiple constraints, enhancing its operation capabilities.