Multi-UAV simultaneous target assignment and path planning based on deep reinforcement learning in dynamic multiple...

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Cooperative multi-unmanned aerial vehicle (UAV) systems require effective target assignment and path planning.
• Environmental dynamics and partial observability pose significant challenges for UAV coordination.

Purpose of the Study:

• To address the challenge of multi-UAV target assignment and path planning in dynamic, partially observable environments.
• To develop a novel algorithm integrating target assignment and path planning for enhanced UAV cooperativity.

Main Methods:

• Formulated the multi-UAV target assignment and path planning problem as a partially observable Markov decision process (POMDP).
• Proposed a deep reinforcement learning (DRL) algorithm incorporating a target assignment network within the twin-delayed deep deterministic policy gradient (TD3) framework.
• The target assignment network handles assignments at each step, while TD3 optimizes path planning and provides training signals.

Main Results:

• The proposed DRL algorithm achieved optimal complete target allocation for UAVs.
• Demonstrated collision-free path planning for each UAV in complex 3D dynamic environments with multiple obstacles.
• The approach exhibited superior performance in target completion and adaptability compared to existing methods.

Conclusions:

• The novel DRL-based algorithm effectively solves the integrated target assignment and path planning problem for multi-UAV systems.
• The method shows significant improvements in cooperative task execution and environmental adaptability.
• This approach offers a robust solution for complex real-world UAV applications.