Target enclosing control of uncrewed surface vehicles (USVs) is essential for maritime monitoring and security. For noncooperative targets in practical scenarios, relying on global information becomes ineffective, thus making vision-based perception a promising solution. However, achieving reliable vision-based enclosing faces significant challenges including field of view (FOV) constraint and various uncertainties. To address these challenges, this article presents a monocular vision-based target enclosing framework for USVs. The proposed framework first incorporates a tailored YOLOv5-based detector coupled with a bearing-only state estimator, which can achieve reliable target state estimation from monocular vision. Then, to maintain continuous visual contact, a control barrier function based safety-critical filter is developed to constrain the USV’s FOV during target enclosing. Furthermore, a dual-layer extended state observer (ESO) based control structure is implemented to handle various uncertainties, where kinematic-level ESOs compensate for unknown USV lateral velocity and target motion, while dynamic-level ESOs address lumped dynamic uncertainties. This approach ensures robust and accurate target enclosing with continuous visual tracking. Both simulations and experiment are implemented to verify the effectiveness of the proposed framework.

Autonomous berthing remains a key challenge for unmanned surface vehicles (USVs), especially in dynamic marine environments where traditional methods relying on GPS, wireless communication, or visual markers face limitations due to signal attenuation and lighting changes. This article proposes a vision-based robust berthing pose estimation framework that integrates infrared light arrays and ArUco fiducial markers, enabling accurate and real-time pose estimation. The YOLO-v5n berth object detection model has been fine-tuned on custom datasets for berthing scenarios, while TensorRT optimization ensures efficient deployment on embedded platforms. The system utilizes adaptive image processing techniques, including Otsu binarization, Canny edge detection, and centroid-based light center localization, to isolate infrared LEDs and extract their geometric features under different illumination conditions. The perspective n-point (PnP) algorithm was used to estimate USV’s relative pose to the light array. The framework has been validated through experiments under different conditions, providing a reliable solution and potential applications for USV autonomous berthing.

Twisted and Coiled Actuators (TCAs) are promising in soft robotics for their high energy density, light weight, and low voltage. However, current TCA-based soft robots face challenges of limited deformation and control precision, mainly due to the preloading requirement of TCAs and the lack of suitable intrinsic sensing capabilities. To address these issues, we designed a TCA with high load capacity, free stroke, and self-sensing capabilities, proposed flexible optical fiber-based posture and tactile sensing methods, and developed a multi-loop feedback controller. Collectively, these enable millimeter-level tracking accuracy in a soft tentacle robot. The TCA, with an optimized manufacturing process, achieves a 30% free stroke without preloading, a 32% improvement in ultimate stress, and temperature self-sensing capabilities with a maximum error of less than 6%. Combining the TCAs with compliant macro-bend optical fibers and soft optical waveguides, we created a soft robotic tentacle with intrinsic posture and tactile sensing, and designed a multi-input multi-output closed-loop and feedforward controller. Experiments demonstrate that the model-based feedforward significantly improves the control performance, reducing the rise time by 15.3%. The trajectory tracking error remains within the millimeter range, and the repetitive positioning error for hexagonal trajectories reaches sub-millimeter precision. The tactile sensor of the robot enables real-time perception of the object's modulus and pressing states. These findings highlight the soft robotic tentacle's potential for various applications, including underwater exploration, detection, and sampling.