Autonomous Path Planning Simulation Method for UAVs in Complex Environments

To address the technical bottlenecks of unmanned aerial vehicle (UAV) autonomous navigation in complex, strongly constrained scenarios such as urban canyons and disaster ruins, this paper proposes a hierarchical decision-making architecture that integrates large-model environmental prior knowledge with small-model real-time control. A dynamic obstacle probability model is constructed through visual - light detection and ranging multi-source data fusion. Combined with a sparse-reward reinforcement learning algorithm inspired by biological ant colonies, dynamic obstacle avoidance and energy-optimal path planning in three-dimensional environments are achieved. Furthermore, within the simulation system for complex scenarios, the limitations of traditional algorithms in planning efficiency and trajectory smoothness in complex terrains are overcome by introducing an improved rapidly-exploring random tree - artificial potential field (RRT-APF) algorithm with terrain gradient constraints and an online knowledge distillation mechanism. This provides a new paradigm for UAV intelligent decision-making in complex environments.