Answers are grounded in this paper and cite their sources.
Urban last-mile delivery with multi-UAV fleets is constrained not only by travel distance, but also by temporary landing-exclusion zones created during parcel drop-off. When several UAVs serve nearby delivery points, these exclusion zones can overlap in space and time, forcing hover waiting and increasing fleet makespan. A further challenge is that rooftop, ground-open, and constrained-corridor deliveries impose different landing radii and occupancy durations, while many scheduling models treat delivery points uniformly. This article proposes an OpenStreetMap (OSM) and large language model (LLM)-augmented spatiotemporal occupancy-aware multi-UAV scheduling framework for smart-city last-mile delivery. OSM building footprints, road networks, and open-space polygons are used to classify delivery points into three landing-zone types, after which representative landing-mode parameters are assigned for exclusion-aware scheduling. An LLM-based adaptive weighting mechanism (gpt-4o-mini-2024-07-18, $T=0$ ) maps zone statistics, conflict density, and urban-morphology cues to priority weights for occupancy-, distance-, and conflict-aware task ordering. Experiments on Tokyo and Shenzhen evaluate 24 configurations across two cities, three fleet sizes, four scheduling strategies, and ten random seeds. The proposed framework reduces makespan by more than 90% relative to a sequential single-trip upper bound reference, confirming the benefit of parallel fleet scheduling under landing-exclusion constraints. Against non-LLM heuristics, the LLM-Adaptive strategy remains competitive and achieves the best result in the resource-constrained Shenzhen 12-UAV case. On a reduced ten-task mixed integer linear programming benchmark solved with CPLEX 22.1 to a gap of at most 0.1%, the LLM-Adaptive strategy closes 4.3%–5.2% of the optimality gap left by the best hand-tuned heuristic at $K=3$ . The results show that OSM-derived zone classification and LLM-based priority weighting provide an interpretable and reproducible approach for exclusion-aware multi-UAV scheduling.
No takes yet. Share an insight, caveat, or question.
Fengjun et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: