Priority-aware task assignment and path planning for efficient and load-balanced multi-UAV operation

Abstract

Unmanned aerial vehicles (UAVs) have emerged as a viable option in numerous real-world applications. In large mission planning systems, demanding the effective completion of a wide variety of tasks, multi-UAV systems are significantly preferred. Collaborative task assignment and path planning have a significant impact on autonomous operation of multi-UAV systems. In this study, we propose a priority-aware task assignment and path planning (AMTP) algorithm based on actor-critic multi-agent reinforcement learning for multi-UAV operations. We utilize an actor-critic multi-agent framework based on centralized training and decentralized execution for task assignment and path planning of heterogeneous UAVs. A multi-agent deep deterministic policy gradient framework is used to train the system to deal with both the elements of the dynamic environment and the heterogeneity of UAVs and tasks. The critic network evaluates the policy by maintaining the global network state information and combines the actions of agents to simplify the training state. On the other hand, agents decide which actions to comply with their local observations to improve the policy. The performance evaluation results obtained through simulation show that the proposed AMTP outperforms existing methods in terms of task completion rate, load balancing, energy efficiency, and convergence time.