Synergistic Hierarchical AI Framework for USV Navigation: Closing the Loop Between Swin-Transformer Perception,...

Area of Science:

• Robotics and Artificial Intelligence
• Ocean Engineering
• Autonomous Systems

Background:

• Autonomous Unmanned Surface Vehicle (USV) operations require advanced navigation, guidance, and control.
• Challenges include sensor-based object detection and energy-aware path planning in complex marine conditions.

Purpose of the Study:

• To propose a novel synergistic AI framework for robust USV autonomy.
• To enhance navigation, guidance, and control for complex ocean engineering scenarios.

Main Methods:

• Integrated Swin-Transformer for semantic risk map generation from visual data.
• Transformer-enhanced A-star (T-ASTAR) for energy-aware static path planning.
• Domain-adapted TD3 agent with energy-aware reward for dynamic path optimization and obstacle avoidance.
• CUDA acceleration for computational efficiency.

Main Results:

• Achieved 30% shorter routes and 70% fewer turns compared to benchmarks.
• Reduced dynamic collisions by 64.7%.
• Demonstrated a 215-fold speed improvement in map generation using CUDA acceleration.

Conclusions:

• The synergistic AI framework significantly advances USV autonomy.
• The approach addresses critical challenges in dynamic environments, object avoidance, and energy-constrained operations.
• Hierarchical integration of AI components is crucial for effective unmanned maritime systems.