A Safe Maritime Path Planning Fusion Algorithm for USVs Based on Reinforcement Learning A* and LSTM-Enhanced DWA

Area of Science:

• Maritime Robotics
• Artificial Intelligence
• Path Planning Algorithms

Background:

• Path planning for Unmanned Surface Vehicles (USVs) in dynamic maritime environments presents significant safety challenges.
• Existing methods struggle with predictability and trajectory smoothness when encountering dynamic obstacles.

Purpose of the Study:

• To develop a reliable hybrid path planning approach for USVs.
• To improve the safety, predictability, and smoothness of USV trajectories in complex marine settings.

Main Methods:

• A hybrid approach combining a reinforcement learning-enhanced A* algorithm and an improved Dynamic Window Approach (DWA).
• A* algorithm enhancements include dynamic neighborhood search, adaptive weighting, and path post-optimization.
• Dynamic obstacle prediction uses a Kalman Filter (KF) integrated with a Long Short-Term Memory (LSTM) network.
• DWA incorporates International Regulations for Preventing Collisions at Sea (COLREGs) for compliant navigation.

Main Results:

• The enhanced A* algorithm demonstrated better adherence to USV kinematic models.
• The improved DWA significantly reduced collision risks in simulations.
• The hybrid approach resulted in shorter path lengths and smoother trajectories.

Conclusions:

• The proposed hybrid path planning method enhances USV navigation safety and efficiency.
• The integration of RL, KF-LSTM, and COLREGs provides a robust solution for dynamic maritime environments.