An Efficient Hierarchical Planning Framework for UAV Navigation in Narrow Dynamic Tunnel

Ground robots are common in tunnels, yet rugged terrain and limited space make UAVs superior due to their agility. However, traditional hierarchical planning methods suffer from low global search efficiency, prolonged local re-planning times, and limited capability for diverse dynamic obstacles. To address these issues, this paper proposes the CG-ST Planner, which integrates coneguided global planning with spatio-temporal detection local re-planning. We design a cone-guided A* algorithm that mitigates the inefficiency of traditional A* by constraining the search space. To avoid dynamic obstacles efficiently, we introduce an event-triggering mechanism based on a moving cylindrical detection area. The UAV focuses solely on this forward region rather than the entire map, significantly reducing the computational burden. Finally, a local window re-planning algorithm ensures rapid responsiveness for obstacle avoidance. Simulation results in various tunnel scenarios demonstrate that our framework improves global planning efficiency by up to 98.57% and reduces local re-planning time by over 90% compared to traditional A* and TEB algorithms. Crucially, in dynamic environments with diverse obstacle trajectories, it maintains a greater safety distance while successfully completing the avoidance task in all simulated scenarios. These findings validate that the CG-ST Planner offers a promising solution for UAV path planning in constrained and dynamic environments.