Dynamic Navigation and Area Assignment of Multiple USVs Based on Multi-Agent Deep Reinforcement Learning.

Area of Science:

• Robotics and Autonomous Systems
• Artificial Intelligence
• Marine Engineering

Background:

• Unmanned Surface Vehicles (USVs) offer autonomous capabilities for maritime tasks.
• Single USV autonomy is limited in dynamic environments with multiple simultaneous objectives.
• Cooperative multi-USV systems are essential for achieving higher success rates in complex missions.

Purpose of the Study:

• To develop a cooperative navigation approach for multiple USVs.
• To enhance the autonomy of USVs in dynamic environments.
• To enable automatic dynamic obstacle avoidance and target area allocation for multiple USVs.

Main Methods:

• A multi-agent deep reinforcement learning (MADRL) approach, specifically the multi-agent deep deterministic policy gradient (MADDPG), was employed.
• The MADDPG framework was utilized to jointly optimize USV trajectories, obstacle avoidance, and coordination.
• A task management system integrating dynamic navigation and area assignment was designed based on the MADDPG framework.

Main Results:

• The proposed method effectively enables multiple USVs to autonomously avoid dynamic obstacles.
• The approach facilitates automatic allocation of target areas among cooperative USVs.
• Experiments on the Gym platform validated the effectiveness of the cooperative navigation strategy.

Conclusions:

• The multi-USV cooperative approach significantly enhances autonomy and task success rates in complex maritime operations.
• Integrating dynamic navigation and area assignment within a MADRL framework offers a robust solution for multi-mission USV deployment.
• The proposed MADDPG-based system demonstrates a promising direction for advanced autonomous maritime systems.