Modeling and Predictive Control of an Automatic Launch and Recycling System for USV Based on NAR Neural Network

Abstract

Abstract Every launch and recycling system of an unmanned surface vehicle (USV) needs to be specially designed for every ship and relies on manual operation to hook and unhook, which results in a high-cost, high-risk, and low success rate. To solve these difficult problems, this study develops a new type of launch and recycle (L&R) system for USVs that realizes automatic guidance docking, launch, and recycling between ships and the USV. The contributions of this paper are mainly: (1) A new type of automatic launch and recycling system with cable-throwing, guiding, and locking mechanisms is proposed to achieve an efficient and automatic physical connection between the USV and ship crane and overcome the problem of the launch and recycling tasks needing manual assistance for the USV. (2) A dynamic predictive approach that combines an estimation of the motion trajectory for the cable-throwing ball (abbreviated as the ball) under multiple disturbances and an attitude predictive model based on a nonlinear autoregressive neural network for USVs is proposed. The proposed approach realizes the accurate prediction of the launch moment for the ball based on the physical connection between the USV and ship and raises the success rate of launching and recycling in harsh environments. Offshore experiments showed that the system requires only 80 s and the success rate is 100% during the launch of a USV; additionally, the system requires no more than 190 s and the success rate is 99% to recycle a USV. Compared to the baseline for L&R method, the proposed method in this paper reduces the application costs and promotes the L&R success rate for the USV. The actual working results show that the developed automatic system is versatile, safe, and highly efficient and meets the requirements of practical applications.