In this work, an improved model predictive visual servo control method with guaranteed stability and feasibility is presented for an underwater vehicle-manipulator system (UVMS) to track the visual reference trajectory of a camera fixed on the end-effector in manipulation tasks. The hybrid visual features including visual moments and quaternions are used, and the new system dynamic equations are derived. A Lyapunov-based model predictive control (LMPC) algorithm is developed for the visual tracking task, which generates the constrained control signals to consistently achieve optimal tracking performance in accordance with the objective function. An auxiliary Lyapunov-based controller is designed to construct the contraction constraint for LMPC. With the assistance of this auxiliary controller and the contraction constraint, the feasibility and stability of the proposed LMPC controller are analyzed. The trouble of the complex local linearization step in traditional model predictive control (MPC) is resolved and the linearization-induced inaccuracy can be avoided. A barrier cost is integrated into the cost function of LMPC to prevent the visual target from escaping the field of view. Finally, comparative simulation experiments are conducted and the results verify the improved convergence and robustness of the proposed high-dimensional visual servo tracking controller of UVMSs.

Robots with multiple degrees of freedom (DOFs), such as underwater vehicle–manipulator systems (UVMSs), are expected to optimize system performance by exploiting redundancy with various basic tasks while still fulfilling the primary objective. Multiple tasks for robots, which are expected to be carried out simultaneously with prescribed priorities, can be referred to as sets of tasks (SOTs). In this work, a hybrid vision/force control method with continuous task transitions is proposed for a UVMS to simultaneously track the reference vision and force trajectory during manipulation. Several tasks with expected objectives and specific priorities are established and combined as SOTs in hybrid vision/force tracking. At different stages, various SOTs are carried out with different emphases. A hierarchical optimization-based whole-body control framework is constructed to obtain the solution in a strictly hierarchical fashion. A continuous transition method is employed to mitigate oscillations during the task switching phase. Finally, comparative simulation experiments are conducted and the results verify the improved convergence of the proposed tracking controller for UVMSs.

Forward-looking sonar (FLS) image segmentation is essential for underwater exploration with remaining challenges including low contrast, ambient noise, and complex backgrounds, which both existing traditional and deep learning-based methods fail to address effectively. This paper presents TriEncoderNet, a novel model that simultaneously extracts local, global, and edge-related features through three parallel encoders. Specifically, the model integrates a convolutional neural network (CNN) for local feature extraction, a transformer for global context modeling, and a histogram of oriented gradients (HOG) encoder for edge and shape detection. The key innovations of TriEncoderNet include the CrossFusionTransformer (CFT) module, which effectively integrates local and global features to capture both fine details and comprehensive context, and the HOG attention gate (HAG) module, which enhances edge detection and preserves semantic consistency across diverse feature types. Additionally, TriEncoderNet introduces the hierarchical efficient transformer (HETransformer) with a lightweight multi-head self-attention mechanism to reduce computational overhead while maintaining global context modeling capability. Experimental results on the marine debris dataset and UATD dataset demonstrate the superior performance of TriEncoderNet. Specifically, it achieves an mIoU of 0.793 and mAP of 0.916 on the marine debris dataset, and an mIoU of 0.582 and mAP of 0.687 on the UATD Dataset, outperforming state-of-the-art methods in both segmentation accuracy and robustness in challenging underwater environments.

This paper investigates a Sarsa-based visual servoing control gain tuning method and the application on a manipulator. For a typical visual servo controller, fixed control gains will not provide the best performance. Therefore, state action reward state action (SARSA) algorithm, one of learning-based methods from reinforcement learning (RL), is introduced to select control gains in every control step. The norm of the visual error is used to define the state space. The positive gain of the controller is discretized as the actions. A reward function is defined to evaluate the performance of every action. Both a numerical test and a robot experiment are carried out to validate the presented algorithm.