DTHV-PPO: A UAV Control Method with Dynamic Task Goal Adaptation and High- Value Experience-Guided Replay Based on PPO

With the rapid development of UAV technology, its control techniques are increasingly applied to complex tasks requiring advanced autonomy. However, existing algorithms are predominantly designed for fixed task target control and often struggle with slower adaptation and instability when faced with dynamically changing task goals. To address these challenges, this paper proposes a reinforcement learning-based control method for fixed-wing UAVs, named Dynamic Task Goal Adaptation and High-Value Experience-Guided Replay Method Based on Proximal Policy Optimization (DTHV-PPO). The method dynamically adjusts task goals during training and prioritizes learning from challenging experiences, thereby enhancing both training efficiency and control per-formance. And it utilizes PPO as the core optimization mechanism, supplemented by an adaptive reward weighting system. Experimental results in scenarios with dynamically changing goals show that DTHV-PPO outperforms JS-PPO, achieving a 7% increase in task success rate and a 2.2% reduction in overshoot.