Research on UAV 3D Path Planning Method Based on Deep Reinforcement Learning

Efficient path planning is crucial for autonomous UAV navigation in dynamic environments. Traditional 3D path planning methods rely on precise environmental models while simultaneously facing challenges in handling high-dimensional complexity and discontinuous speed control. To overcome these challenges, we propose a UAV path planning approach for 3D obstacle spaces based on the Proximal Policy Optimization (PPO) algorithm. Our method integrates a recurrent neural network into the PPO framework to process time-series data effectively, enabling the UAV to leverage historical information for improved decision-making in dynamic scenarios. The exploration is further enhanced by designing an environmental reward scheme that combines angle-based rewards with a curiosity module, which helps the UAV rapidly discover optimal paths. To ensure a smooth flight trajectory, the planned path is optimized using the Minimum Snap method. The overall approach is validated on the UE4+AirSim platform demonstrating its practical applicability for real-world UAV flight decision-making. The experimental results demonstrate the superiority of this method.