Perception-Aware Cooperative Path Planning for Multi-UAV Systems in Urban Wind Fields via Deep Reinforcement Learning

Area of Science:

• Robotics
• Artificial Intelligence
• Aerospace Engineering

Background:

• Multi-UAV operations in urban environments face challenges from complex structures and wind.
• Environmental disturbances degrade UAV trajectory accuracy and policy robustness.

Purpose of the Study:

• To develop a perception-aware cooperative path planning algorithm for multi-UAVs in challenging urban environments.
• To enhance the robustness and efficiency of multi-UAV operations under wind disturbances.

Main Methods:

• An enhanced Dueling Double Deep Q-Network (D3QN) algorithm, NPD3QN, was proposed.
• Environmental data was formulated into a Markov Decision Process with an N-step update strategy.
• An improved Prioritized Experience Replay (PER) mechanism was implemented for training stability.

Main Results:

• NPD3QN effectively maps high-dimensional perceptions to robust control commands.
• In simulated wind, NPD3QN reduced total path length by 11.7% compared to standard D3QN.
• The algorithm generated streamlined cooperative trajectories in wind-disturbed scenarios.

Conclusions:

• NPD3QN provides a robust, sensor-driven foundation for autonomous multi-UAV path planning.
• The method enhances multi-UAV performance in complex, wind-affected urban airspaces.
• Further evaluation in real-world conditions is warranted.