Path planning for UAVs in complex terrain based on the PGD model: Algorithmic improvements combining feature...

Area of Science:

• Robotics and Artificial Intelligence
• Autonomous Systems
• Computational Intelligence

Background:

• UAV path planning in complex terrains faces challenges with high-dimensional data, inefficient exploration, and limited adaptability.
• Existing methods struggle to process complex environmental data and generate robust, adaptable flight paths.

Purpose of the Study:

• To propose the PGD model, a novel framework for efficient and adaptive UAV path planning in complex terrains.
• To address limitations in high-dimensional state processing, blind path exploration, and cross-scene adaptability.

Main Methods:

• The PGD model employs a synergistic approach, integrating Transformer for data compression, Generative Adversarial Network (GAN) for path generation, and Deep Deterministic Policy Gradient (DDPG) for strategy optimization.
• This creates a closed-loop system for "compression-generation-optimization" in path planning.

Main Results:

• PGD demonstrated superior performance on UAVDT and AirSim datasets, achieving shorter path lengths (20.0m/22.0m) compared to baselines (23.8m/24.0m).
• PGD significantly reduced collision rates (2.5%/3.0%) and improved computational efficiency (13.5s/16.0s).
• The model showed enhanced feature correlation and physical path constraints through multi-module synergy.

Conclusions:

• The PGD model offers significant improvements in UAV path planning efficiency and adaptability, especially in high-complexity environments.
• Its novel framework provides a robust solution for intelligent navigation in challenging terrains.
• Future research will explore adaptability to extreme weather and multi-agent scenarios.