Dynamic Reward-Based Deep Reinforcement Learning Algorithm for UAV Path Planning in Large-Scale Environments

Path planning for Unmanned Aerial Vehicles (UAV) is a vital component of navigation in robotics. The reinforcement Q-learning algorithm enhances path planning for drones but suffers from the need for a large Q-value table and challenges in complex navigation situations. By integrating deep learning with reinforcement one, these shortcomings can be addressed. In this paper, a Deep Q-Network (DQN) model is developed and trained to estimate the drone’s state-action value function. In this work, the flight space is represented by a grid of cells, which are then encoded to convert environmental information into a new input format suitable for the DQN model. Normalizing state inputs enhances the stability and convergence of the proposed DQN algorithm by ensuring comparability of features across different scales. Besides, a new dynamic reward function is established based on the distance between the drone’s current position and its destination. Simulation results and discussion illustrate the effectiveness of the proposed DQN-based approach for collision-free path planning of UAV in complex environments.