Energy-Efficient Online Path Planning for Internet of Drones Using Reinforcement Learning

Unmanned aerial vehicles (UAVs) have recently been applied in several contexts due to their flexibility, mobility, and fast deployment. One of the essential aspects of multi-UAV systems is path planning, which autonomously determines paths for drones from starting points to destination points. However, UAVs face many obstacles in their routes, potentially causing loss or damage. Several heuristic approaches have been investigated to address collision avoidance. These approaches are generally applied in static environments where the environment is known in advance and paths are generated offline, making them unsuitable for unknown or dynamic environments. Additionally, limited flight times due to battery constraints pose another challenge in multi-UAV path planning. Reinforcement learning (RL) emerges as a promising candidate to generate collision-free paths for drones in dynamic environments due to its adaptability and generalization capabilities. In this study, we propose a framework to provide a novel solution for multi-UAV path planning in a 3D dynamic environment. The improved particle swarm optimization with reinforcement learning (IPSO-RL) framework is designed to tackle the multi-UAV path planning problem in a fully distributed and reactive manner. The framework integrates IPSO with deep RL to provide the drone with additional feedback and guidance to operate more sustainably. This integration incorporates a unique reward system that can adapt to various environments. Simulations demonstrate the effectiveness of the IPSO-RL approach, showing superior results in terms of collision avoidance, path length, and energy efficiency compared to other benchmarks. The results also illustrate that the proposed IPSO-RL framework can acquire a feasible and effective route successfully with minimum energy consumption in complicated environments.