An Autonomous Path Planning Model for Unmanned Ships Based on Deep Reinforcement Learning

Area of Science:

• Marine Engineering
• Artificial Intelligence
• Robotics

Background:

• Deep reinforcement learning (DRL) excels in continuous control tasks, finding applications in path planning.
• Intelligent path planning for unmanned ships in unknown environments remains a challenge.

Purpose of the Study:

• To develop an autonomous path planning model for unmanned ships using DRL.
• To enhance the safety and accuracy of planned paths by incorporating navigation rules and restricted areas.
• To improve the convergence speed and stability of DRL algorithms in maritime applications.

Main Methods:

• Utilized the deep deterministic policy gradient (DDPG) algorithm for optimal action strategy learning.
• Integrated ship data from the Automatic Identification System (AIS) for model training.
• Combined DDPG with the artificial potential field method for an improved DRL approach.
• Implemented and tested the path planning model on an electronic chart platform.

Main Results:

• The developed DRL model successfully achieved autonomous path planning for unmanned ships.
• The improved DRL model demonstrated good convergence speed and stability in experimental results.
• The integration of navigation rules and encounter situations into restricted areas ensured path safety.

Conclusions:

• The proposed DRL-based autonomous path planning model is effective for unmanned ships.
• The hybrid DDPG and artificial potential field approach offers enhanced performance.
• The model shows promise for safe and efficient maritime autonomous navigation.