Dr. Deng Yingjie

Currently, I am focusing on the path tracking problem of unmanned ships under ocean disturbances and have established an integrated control framework of 'guidance—heading control—learning-based disturbance estimation compensation.' The upper layer uses variable LOS to generate reference heading, while the lower layer constructs a heading controller with prescribed performance and fixed-time characteristics, with a focus on designing a learning-based total disturbance estimator. A 'two-stage learning' approach is adopted: first, supervised pre-training of the total disturbance based on simulation data, followed by fine-tuning the estimator with reinforcement learning. Current results indicate that the learning-based disturbance estimator can already stably improve tracking performance in the pure heading channel, and after RL fine-tuning, it further reduces heading errors and control costs compared to purely supervised estimation. At the path tracking level, three comparative sets—without compensation, ideal compensation, and learning-based compensation—have been completed.