Autonomous flight is a critical capability for unmanned aerial vehicles (UAVs), enabling applications in wildlife and plant protection, infrastructure inspection, search and rescue, and other complex missions. Although some learning-based methods have achieved considerable progress, traditional algorithms still struggle with real-world challenges, due to the partially observable nature of environments and limited experience regarding the properties of dynamic unknown environments where threats and targets are movable and unpredictable. To address these difficulties, it is necessary to achieve autonomous guidance for UAVs performing long-range missions in dynamic environments (LRGDEs), and to develop a novel end-to-end algorithm that can overcome partial observability under limited state transitions. In this paper, we propose an RNN-enhanced Diverse Curriculum-driven Learning Algorithm (REDCRL) based on deep reinforcement learning. We modify the structure of traditional actor–critic networks and introduce Bi-LSTM into policy networks (referred to as Bi-LSTM-modified Policy Networks (BLPNs)) to alleviate observation incompleteness. Furthermore, to fully exploit the potential value of data and mitigate the problem of insufficient samples, we develop an Adaptive Multi-Feature Evaluation Experience Replay (AMFER) method to reshape the process of experience replay buffer construction and sampling. In addition, the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm is adopted to optimize UAV-maneuver decision policies. Compared with traditional algorithms, the proposed algorithm can accelerate policy convergence and improve the performance of the trained policy.

Multi-UAV multi-sensor cooperative detection is critical for situational awareness in complex environments with stationary known targets. To overcome the shortcomings of existing task allocation models in fine-grained cooperation, stealth constraints, and large-scale optimization, this paper proposes a new offline mission pre-planning method for static target scenarios. A five-layer hierarchical resource scheduling model is established, covering task, UAV platform, sensor type, cooperative mode, and execution role. Stealth requirements are incorporated as hard constraints, and a bi-objective optimization model is formulated to maximize weighted detection efficacy. A statistical adaptive genetic algorithm (SAGA) is developed as an offline pre-planning algorithm, featuring a coupled chromosome encoding strategy that integrates the five decision layers and an adaptive crossover-mutation mechanism guided by the Wilcoxon signed-rank test with effect size monitoring. Simulation results across small, medium, and large scenarios show that the SAGA achieves the highest mean best fitness in all scenarios. Compared with the GA, the improvement is statistically significant in the large-scale scenario, while in small- and medium-scale scenarios the differences from the GA and AGA are not significant. Compared with an improved whale optimization algorithm, the SAGA achieves statistically significantly higher best fitness and shorter computational time in all scenarios and converges significantly faster in small- and medium-scale scenarios. However, the SAGA requires longer computation time than the GA and AGA due to the per-generation statistical test. All allocations strictly satisfy the prescribed stealth constraints in the simulation model. The proposed method provides a novel, quantitatively validated solution for fine-grained, constraint-aware, and computationally efficient multi-UAV multi-sensor cooperative detection mission pre-planning.