Towards Automated and Sustainable Airport Surface Movement Operations

Abstract

Airside operations at major airports are under increasing pressure as traffic demand grows, environmental constraints tighten, and expectations for safety remain uncompromised. On the airport surface, increasing congestion, uncertain taxiing behaviour, and rising operational complexity place additional demands on Air Traffic Control Officers (ATCOs), intensifying the risk of inefficiencies, elevated workload, and human error. In parallel, the transition towards more sustainable ground operations, such as the emergence of engine-off taxiing techniques, require a fundamental rethinking of the existing airport operations.

Against this backdrop, regulators, researchers, and practitioners increasingly view AI-enabled decision-support and automation as key enablers for future improvements. Yet, the literature lacks a modelling framework that simultaneously captures the interrelated operational processes, heterogeneous actors, and fidelity requirements necessary to evaluate next-generation concepts for airport surface movement operations (ASM Ops) in a realistic and systematic manner. Many existing models remain limited to isolated subsystems, single concepts, or strong simplifying assumptions, which restrict their ability to compare alternative future approaches at a detailed level. To address this gap, the objective of this thesis is to design models, methods, and tools to investigate next-generation concepts of operations (ConOps) for automated and sustainable ASM Ops.

The thesis is structured around eight research questions (RQs) and proceeds in three phases. The development phase (Chapters 2 to 5) first derives modelling requirements from the operational structure of ASM Ops and from the goal to represent surface movements at high fidelity while remaining computationally tractable (RQ 1). ASM Ops encompasses a hierarchy of tasks, ranging from strategic planning (long-term and high-level decisions) to scheduling (tactical allocation of resources), routing (tactical planning of movements), guidance (operational planning of movements), and finally movement execution (controlling aircraft and ground vehicles). This hierarchical decomposition, combined with the need for coordinated oversight to maintain the strict safety standards in aviation, motivates a hierarchical–distributed modelling approach.

To provide the required modularity, adaptability, and expressiveness in both fast-time and real-time simulations, this thesis adopts the multi-agent systems (MAS) paradigm. A generalised MAS architecture is proposed, comprising environmental objects and four agent categories aligned with the hierarchy of ASM Ops tasks (RQ 2). The architecture supports different allocations of responsibilities and different coordination mechanisms, enabling the instantiation of tailored model instances for specific ConOps while retaining a reusable simulation environment and component structure.

To support trajectory-based automation studies at airport scale (RQ 3), the thesis then outlines the Multi-Agent Motion Planning on Airport Surfaces (AS-MAMP) algorithm as the decision-logic for centralised path planning in a fully-automated operational setting, i.e. on EASA’s AI Level 3. AS-MAMP is a two-level solver that builds on Priority-Based Search (PBS) and its variants for high-level conflict resolution. Because existing low-level planners were insufficient to compute realistic 4D ground trajectories under operational constraints, the thesis introduces the novel Safe Interval Motion Planning (SIMP) algorithm. SIMP plans continuous-time trajectories in accordance with the operational processes in ASM Ops (e.g. pushback and engine-start, tug coupling/decoupling, and holding), motions based on finite acceleration, and conflict avoidance in continuous space and time.

The resulting planning approach is evaluated through structured benchmarking on a synthetic airport layout, where the high-level coordination component is compared against PBS variants and SIMP is benchmarked against SIPP and kinodynamic A*. The evaluation is complemented by experiments on the real-world layout of Amsterdam Airport Schiphol to assess scalability and operational relevance under realistic airport complexity. Taken together, these results establish the MAS architecture and AS-MAMP planning capability as an enabling foundation for subsequent operational analyses, while also identifying limitations and implementation challenges that arise when moving towards real-world deployment.

The validation phase (Chapters 6 to 8) applies the developed MAS framework in operational studies and subsequently strengthens its realism. Two far-term analyses investigate fully-automated ASM Ops under multi-engine taxiing (RQ 4) and under engine-off taxiing (RQ 5), examining system-level implications under dense traffic assumptions. Building on feedback from ATCOs and other operational experts gathered throughout the thesis, historical aircraft ground tracks are then analysed to improve the calibration of key movement parameters (RQ 6) so that the model is able to more accurately reproduce the historical operations. The MAS model is further extended to represent aircraft towing movements and to assess their impact on regular traffic (RQ 7). Based on these modifications, multiple operational model variations are compared to clarify the sensitivity of results to modelling assumptions and to discuss implications for next-generation surface operations.

Fully-automated operations pose significant implementation challenges in real-world settings, so human involvement remains required for the foreseeable future. Therefore, as a final exploration phase, Chapter 9 provides an outlook towards EASA’s AI Level 2 concepts by examining how human–automation interaction mechanisms can be embedded into the MAS model to enable operator involvement (RQ 8). These demonstrations focus on technical feasibility – illustrating how interfaces and information exchange could be embedded into the MAS model – rather than validated evidence from human-in-the-loop experiments. Chapter 10 concludes by synthesising the findings across all research questions, reflecting on limitations and implications for future ASM Ops, and outlining directions for continued development and socio-technical validation.