A Multi-AUV Maritime Target Search Method for Moving and Invisible Objects Based on Multi-Agent Deep Reinforcement...

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Oceanography

Background:

• Searching for moving and invisible objects in dynamic ocean environments presents significant challenges.
• Traditional search methods struggle with the unpredictable drift of targets in water currents.

Purpose of the Study:

• To develop and evaluate a novel multi-agent target search method (MATSMI) for autonomous underwater vehicles (AUVs).
• To enhance the efficiency and success rate of maritime target search operations.

Main Methods:

• The study proposes the Multi-Agent Target Search Method with Information (MATSMI) algorithm, an advancement of the multi-agent deep deterministic policy gradient (MADDPG).
• MATSMI incorporates spatial and temporal information into the reinforcement learning state and defines scenario-specific rewards.
• A simulation environment was developed to model multi-AUV search scenarios for floating objects.

Main Results:

• The MATSMI method demonstrated a 20% higher search success rate compared to traditional search techniques.
• MATSMI achieved search completion in approximately 70 fewer steps.
• The algorithm exhibited faster convergence rates than the standard MADDPG method.

Conclusions:

• MATSMI offers a novel and effective solution for the complex problem of maritime target search.
• The method shows significant improvements in both efficiency and success rate for AUV-based search operations.
• This research contributes a valuable tool for future underwater exploration and recovery missions.