USV Formation and Path-Following Control via Deep Reinforcement Learning With Random Braking

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Marine Engineering

Background:

• Unmanned Surface Vessels (USVs) require robust formation control for coordinated tasks.
• Path following in USV formations presents challenges due to underactuation and dynamic environments.
• Existing deep reinforcement learning (DRL) methods can face local optima during training.

Purpose of the Study:

• To develop an advanced formation control strategy for USVs capable of path following.
• To enhance the adaptability and robustness of USV formations using a novel DRL approach.
• To address the limitations of traditional DRL in achieving optimal formation control objectives.

Main Methods:

• A modified deep reinforcement learning with random braking (DRLRB) algorithm was developed.
• A formation control model using DRL was constructed, incorporating velocity and error distance rewards.
• A random braking mechanism was introduced to prevent training from local optima.
• A virtual leader-based path-following guidance system was implemented for the USV formation.

Main Results:

• The DRLRB method successfully enabled USVs to form and maintain preset formations.
• The system demonstrated automatic and flexible formation adjustments, even when individual USVs deviated.
• Simulations confirmed the effectiveness and superiority of the proposed formation and path-following control strategy.
• The random braking mechanism improved the training stability and convergence of the decision-making network.

Conclusions:

• The proposed DRLRB strategy offers a significant advancement in USV formation path following.
• The method provides robust and adaptive control for multi-USV systems in complex scenarios.
• This research contributes to the development of intelligent autonomous marine systems.