A Deep Reinforcement Learning Approach for UAV Path Planning Incorporating Vehicle Dynamics with Acceleration Control.

This article presents a novel deep reinforcement learning (DRL) approach for unmanned aerial vehicle (UAV) path planning that incorporates vehicle dynamics through acceleration-based control in a continuous three-dimensional (3D) environment. Utilizing the Deep Deterministic Policy Gradient (DDPG) algorithm, the model accounts for UAV velocity, altitude changes, and drag force to more accurately reflect real-world flight conditions. A new reward function based on the inner product is introduced to simultaneously guide target tracking and obstacle avoidance maneuvers. Training and simulation results demonstrate that the UAV effectively learns to navigate complex environments, achieving high success rates even in congested scenarios, and outperforming comparable discrete-action and traditional reward function methods. The study highlights the potential of acceleration control and continuous state-action spaces in enhancing UAV autonomous navigation and suggests future work to include more complex environmental factors such as wind and varied obstacle geometries.