Learning-Based Guidance and Control Codesign for Underactuated Autonomous Surface Vehicles: Theory and Experiment
Li‐Ying Hao, Yun-Peng Liu, Zhi‐Jie Wu, Chao Shen
Abstract
Traditional sideslip angle estimator compensation methods used in line-of-sight (LOS) guidance law are not effective when facing with big amplitude and fluctuating sideslip angle scenarios, resulting in poor path following performance of underactuated autonomous surface vehicles (ASVs). To overcome the drawback, a novel sideslip angle estimator based on long short-term memory (LSTM) is proposed in this article. It integrates the selective updated strategy (SUS) to enhance the learning capability and long-term memory of extremely fluctuating temporal information, thereby meeting the requirement of estimating fluctuating sideslip angle. Based on the proposed SUS-LSTM sideslip angle estimator, a learning line-of-sight (LLOS) guidance law for path following is designed. Furthermore, we theoretically prove the input-to-state stability in probability of the closed-loop cascaded control system, which consists of LLOS and heading controller. Finally, the proposed algorithm is implemented on ASVs and experiments are conducted in the Lingshui bay to validate the superiority and effectiveness of the algorithm.