This research introduces a novel approach for training Unmanned Surface Vehicle (USV) swarms by incorporating human feedback into multi-agent reinforcement learning systems.

• Addresses the challenge of aligning agent behavior with human intentions in complex security operations
• Implements Reinforcement Learning with Human Feedback (RLHF) techniques adapted for multi-agent systems
• Demonstrates improved performance in search and rescue, surveillance, and vessel protection scenarios
• Creates more intuitive and human-aligned behavior patterns without explicitly coding all preferences

This advancement has significant implications for maritime security operations where automated systems must balance operational effectiveness with human tactical preferences and intuition.

Human Implicit Preference-Based Policy Fine-tuning for Multi-Agent Reinforcement Learning in USV Swarm