强化学习
计算机科学
人工智能
搜救
工程类
运筹学
机器学习
随机过程
马尔可夫决策过程
风浪
随机规划
作者
Yutong Song,Tianyi Zeng,Yao Zhang,Tahsin Tezdogan
标识
DOI:10.1016/j.oceaneng.2026.125408
摘要
Maritime search and rescue (SAR) in stochastic wave environments presents a critical challenge for multi-Unmanned Surface Vehicle (USV) systems and demands a fine balance between search efficiency and operational safety. This paper proposes a novel hierarchical reinforcement learning framework, termed Non-Causal Reward Multi-Agent Proximal Policy Optimization (NCR-MAPPO), to address this challenge. Our framework decouples the mission by employing a strategic guidance system based on International Maritime Organization (IMO) standards for systematic coverage while a tactical motion controller built upon the Multi-Agent Proximal Policy Optimization (MAPPO) algorithm learns cooperative execution. The core innovation is a Non-Causal Reward (NCR) mechanism that incorporates short-term wave field prediction into the decision process, enabling a shift from reactive collision avoidance to proactive seakeeping control. Through comprehensive simulations, we demonstrate the superiority of our framework. Compared to the standard MAPPO baseline, NCR-MAPPO significantly enhances survivability by reducing wave impact incidents by 27% and exposure to hazardous sea states by 25% while maintaining high mission efficiency. This work provides a robust solution for autonomous marine systems by bridging the gap between regulatory compliance and predictive safety control.
科研通智能强力驱动
Strongly Powered by AbleSci AI