New HTNav framework combines imitation and reinforcement learning to improve aerial drone navigation in complex urban environments

HTNav addresses key challenges in aerial Vision-and-Language Navigation (VLN) including poor generalization to unseen scenes and inadequate spatial continuity understanding

Hybrid IL-RL framework uses staged training mechanism to balance stability of basic navigation strategy with enhanced environmental exploration capabilities

Tiered decision-making mechanism enables collaborative interaction between macro-level path planning and fine-grained navigation control

Designed for practical applications like logistics delivery and urban inspection requiring long-range path planning in complex urban settings