Semantic-Constrained Planning for Airport Vehicle Scheduling
Sheng Wang, Tianhe Chi · Applied Sciences · 2026
As airport operations expand and ground handling becomes more complex, airport vehicle scheduling has evolved into a system-level decision problem constrained by operational rules, task dependencies, and resource availability. However, existing approaches largely rely on statistical correlation modeling and lack explicit representations of operational semantics and feasibility constraints, resulting in limited executability and poor cross-scenario robustness. To address this issue, we propose the Semantic-Constrained Planning Network (SCP-Net), which adopts a compile-first, plan-later paradigm by embedding operational semantics directly into the scheduling process. SCP-Net introduces an Operational Semantic Compiler (OSC) that encodes key flight task attributes, including service types, operational phases, and time windows, into a structured dependency representation, explicitly modeling task dependencies and task–vehicle feasibility relations. Based on this representation, a Constraint-Gated Planner (CGP) integrates operational dependencies and resource constraints through feasibility-aware gating, ensuring that planning is always conducted within valid operational regions. Through this design, SCP-Net directly generates schedules that are structurally consistent, semantically valid, and executable. Experimental results demonstrate that SCP-Net outperforms baseline methods in terms of executability, constraint violation rate, and cross-scenario stability, highlighting the effectiveness of explicit semantic modeling and constraint-driven planning for airport vehicle scheduling.