Aircraft Turnaround Prediction, GSE Fleet Coordination, and Operational Integration for Airside Autonomous Vehicles

1. Aircraft Turnaround Phases

1.1 Standard Turnaround Sequence

An aircraft turnaround is the complete ground handling cycle between "chocks on" (arrival at gate) and "chocks off" (pushback for departure). The canonical sequence proceeds as follows:

Phase 0 — Arrival and Parking

Aircraft taxis to stand, guided by marshaller or VDGS (Visual Docking Guidance System)
Engines shut down, chocks placed, ground power connected
Typical duration: 2–4 minutes from stop to chocks-on

Phase 1 — Opening and Initial Access (3–4 minutes)

Jet bridge or air stairs positioned and doors opened
Cargo doors opened, belt loaders and container loaders positioned
Ground power unit (GPU) and pre-conditioned air (PCA) connected

Phase 2 — Unloading (8–15 minutes, parallel)

Passenger deplaning: ~8 min for 160 pax (narrowbody), 15–25 min for widebody
Cargo/baggage unloading: ~12 min for 8 ULD containers (narrowbody)
Bulk cargo unloading: ~4 min for 1,200 lbs
Wastewater servicing begins simultaneously

Phase 3 — Servicing (15–25 minutes, parallel, critical path)

Cabin cleaning: 8–15 min (narrowbody), 30–45 min (widebody full clean)
Catering: ~10–13.5 min per galley exchange
Refueling: ~20 min for 26,000 liters (narrowbody), 30–45 min (widebody)
Potable water replenishment
Lavatory servicing
Refueling is typically the critical-path activity in this phase

Phase 4 — Loading (12–15 minutes, parallel)

Cargo/baggage loading: ~12 min for 8 ULD containers
Passenger boarding: ~13 min for 160 pax (narrowbody), 25–40 min (widebody)
Final cargo reconciliation and load sheet preparation

Phase 5 — Departure Preparation (3–5 minutes)

Doors closed, jet bridge retracted
All GSE removed from safety zone
Pushback tug connected
Final headcount verification and safety checks
ATC clearance requested

Phase 6 — Pushback and Taxi

Chocks removed (chocks off = official departure time)
Aircraft pushed back from stand
Engine start during or after pushback

1.2 Typical Durations by Aircraft Type

Aircraft Type	Category	Planned Turnaround	Typical Actual	Key Constraint
A320 / B737	Narrowbody	35–45 min	50–70 min	Refueling (~20 min)
A321neo / B737 MAX 10	Large Narrowbody	40–50 min	55–75 min	Boarding (~15 min)
B787 / A330	Small Widebody	75–90 min	90–120 min	Cargo + Catering
B777	Large Widebody	90 min	105–120 min	Multi-galley catering
A380	Super Heavy	90–105 min	105–140 min	Dual-deck boarding/cleaning

Low-cost carriers (Ryanair, Spirit, AirAsia) routinely achieve 25–30 minute narrowbody turnarounds by eliminating catering exchanges, minimizing cleaning scope, and using single-aisle rapid boarding. Full-service carriers typically plan 45–65 minutes for narrowbody domestic turns.

OAG data from March 2023 shows actual performance variance:

Southwest: 52 min planned, 59 min actual (13% variance)
United Airlines: 62 min planned, 68 min actual (10% variance)
JetBlue: 63 min planned, 75 min actual (19% variance)
B777 turnarounds showed the worst variance: American Airlines experienced 56% variance from plan

1.3 IATA AHM Ground Handling Standards

The IATA Airport Handling Manual (AHM), now in its 46th edition (2026), establishes policy-level standards for passenger, cargo, mail, and aircraft handling, load control, safety management, ground handling agreements, GSE management, and training. The AHM defines "what to do."

The IATA Ground Operations Manual (IGOM), now in its 14th edition (2026), is the complementary procedural manual defining "how to do" — standardized procedures for turnaround actions, chocking, safety briefings, special cargo handling, and passenger processes.

Key IATA standards relevant to turnaround coordination:

AHM Chapter 610: Ground handling agreements (Standard Ground Handling Agreement — SGHA)
AHM Chapter 810: GSE specifications and management requirements
IGOM Chapter 4: Aircraft turnaround actions, standardized chocking procedures, safety briefings
IATA Enhanced GSE Recognition Program: Launched 2024, 98 ground handling fleets registered and 28 stations recognized as of May 2025
IATA Autonomous GSE Guidelines: Published recommended practices for testing and implementing autonomous GSE, addressing unique challenges of the ramp operating environment

2. Turnaround Prediction

2.1 Predicting Phase Transitions

Turnaround prediction aims to forecast when each sub-process will begin and end, culminating in accurate prediction of the Target Off-Block Time (TOBT). This is critical for Airport Collaborative Decision Making (A-CDM), where accurate TOBT feeds into the pre-departure sequencer to generate Target Start-up Approval Time (TSAT) and Target Take-Off Time (TTOT).

The challenge is inherently sequential-stochastic: turnaround sub-processes are partially ordered, partially parallel, and each has variable duration depending on aircraft type, load factor, ground handler performance, weather, and upstream delays.

Key prediction targets:

Phase transition timestamps (when does unloading end? when does boarding begin?)
TOBT / Predicted Off-Block Time (POBT)
Predicted Readiness Time (PRDT)
Total turnaround time
Delay propagation risk

2.2 ML Approaches

Cascaded Gradient Boosting (CGBRT) A cascaded multi-output Gradient Boosting Regression Tree model dynamically predicts aircraft turnaround milestone times. The cascaded framework incorporates hierarchical information transmission — predictions from earlier milestones feed as features into later milestone predictions. Published results (2025) show initial prediction accuracy above 80% within +/- 5 minutes, improving progressively as turnaround operations advance, with over 60% of activities ultimately attaining prediction accuracy above 95% within the same threshold. The model uses flight-related attributes combined with hierarchical features from preceding milestone predictions.

Turnaround Sub-Process Fusion Models Data-driven fusion models integrate sequential information from the turnaround pattern, considering duration of various sub-processes and their overlapping conditions. Machine learning classification algorithms predict turnaround time incrementally based on real-life data, with results showing enhanced robustness and reliability compared to single-output models.

Time Transition Petri Net (TTPN) with Bayesian Inference A hybrid approach models turnaround operations as a Petri net with 12 transitions (landing through departure), then applies Bayesian inference with Monte Carlo simulation for probabilistic duration prediction. Using data from a major Chinese airport (2023), the model achieved RMSE of 3.75 minutes and MAE of 3.40 minutes across both A320 (C-type) and B777 (D-type) aircraft.

Probabilistic and Ensemble Methods Research across Prague, Geneva, Arlanda, and Fiumicino airports demonstrated that linear regression and elastic nets are effective for turnaround time prediction within the A-CDM framework. Deep learning ensembles combining LSTM and gradient boosting have shown high accuracy, particularly when trained on airport-specific operational data.

Graph Neural Networks for Temporal-Spatial Modeling

Graph Attention Networks stacked with LSTM (GAT-LSTM) capture both spatial dependencies between gates/stands and temporal correlations in turnaround sequences
Adaptive Airport Graph Neural Networks (AAGNN) model flight sequence prediction for pre-tactical air traffic management
Dynamic Spatial-Temporal Propagation Neural Networks model delay propagation across airport networks
Spatiotemporal Propagation Networks (STPN) using space-time separable graph convolutions capture how delays from turnarounds propagate to downstream flights

Agent-Based Simulation Agent-based models simulate aircraft stand operations where individual GSE vehicles, crew members, and aircraft act as agents with defined behaviors, producing ground time predictions through emergent system behavior rather than direct regression.

2.3 Data Sources

A-CDM Milestones The A-CDM framework defines 16 milestones tracking flight progress from initial planning to takeoff. Key timestamps:

Timestamp	Definition
ELDT	Estimated Landing Time — predicted touchdown
ALDT	Actual Landing Time — observed touchdown
EIBT	Estimated In-Block Time — predicted arrival at stand
AIBT	Actual In-Block Time — observed chocks-on
TOBT	Target Off-Block Time — predicted readiness for pushback
TSAT	Target Start-up Approval Time — ATC-allocated startup slot
AOBT	Actual Off-Block Time — observed pushback
TTOT	Target Take-Off Time — planned takeoff considering TOBT + taxi
ATOT	Actual Take-Off Time — observed departure

The dynamic interplay between TOBT prediction and TSAT allocation forms the core mechanism of A-CDM. Improved TOBT accuracy directly translates to better runway sequencing and reduced taxi-out delays.

Airport Operational Database (AODB) The AODB serves as the airport's central nervous system — a centralized database storing and distributing real-time operational data including flight schedules, gate assignments, baggage system status, and resource allocations. Key integrations:

Flight Information Display Systems (FIDS)
Baggage Handling Systems (BHS)
Resource Management Systems (stand allocation, check-in desks)
Ground handler dispatch systems
Air Traffic Control data feeds

Major AODB vendors include Amadeus, TAV Technologies, Veovo, Collins Aerospace, and Framfor. The AODB provides the foundational data layer that turnaround prediction models consume: scheduled times, actual times, aircraft type, airline, origin/destination, gate assignment, and historical performance.

ADS-B (Automatic Dependent Surveillance-Broadcast) ADS-B Out broadcasts aircraft GPS position, altitude, ground speed, and identification once per second. For turnaround prediction, ADS-B provides:

Accurate Estimated Landing Time (ELDT) from approach trajectory
Taxi-in time estimation from runway to gate
Real-time awareness of inbound aircraft for proactive GSE dispatch
Airport Surface Detection (ASSC/ASDE-X) fuses ADS-B with multilateration to track aircraft and equipped ground vehicles on the surface movement area

Computer Vision (Apron Cameras) Assaia's ApronAI system exemplifies camera-based turnaround monitoring. Strategically placed cameras across aircraft stands, aerobridges, and the apron generate real-time timestamps for turnaround events (door open/close, GSE arrival/departure, boarding start/end). Deployments include:

Heathrow: 540+ cameras across 116 gates at Terminals 2, 3, and 5
Calgary: 67 gates
JFKIAT, Rome Fiumicino, Toronto Pearson

Performance results from Assaia deployments:

5-minute reduction in ground delays at JFKIAT
17% on-time performance improvement for Alaska Airlines
6-minute delay reduction at Rome Fiumicino
44% reduction in average taxi-in time at Toronto Pearson

3. Moonware HALO Platform

3.1 Overview and Architecture

Moonware, a Los Angeles-based company, has developed HALO — the world's first AI-powered Ground Traffic Control (GTC) platform. HALO serves as a centralized airside operating system that coordinates, monitors, and manages all ground operations in real time.

Core Architecture Components:

Centralized Coordination Engine: Algorithmically dispatches crew and equipment by considering distance, departure/arrival times, crew availability, and real-time schedule changes
Three Data Streams: (1) Real-time flight information from airline and airport systems, (2) crew schedules and task allocation data, (3) ground positions and movement of crew and vehicles via cell phone GPS and low-cost trackers
Dynamic Task Allocation: Automatically assigns and reassigns tasks based on real-time schedule changes, logging precise start/stop times
Live Operational Map: Real-time visualization of aircraft, equipment, and crew locations providing complete operational picture
Communication Layer: Replaces walkie-talkie coordination with digital real-time communication between ground handling control stations and field teams

Key Capabilities:

Dynamic dispatching with live status updates
End-to-end visibility of turnaround operations
Timestamped activity tracking fed continuously into planning tools
Data-driven decision support for long-term process improvements
Adaptation to fluctuating schedule changes for swift resource allocation

3.2 Performance Claims

20% reduction in delays across deployed stations
5-minute average decrease in turnaround time
Reduced aircraft downtime and optimized gate utilization
Faster turnarounds, reduced block times, higher flight throughput, and maximized asset utility

3.3 Deployments

JFK Terminal 8 — British Airways / IAG Moonware's inaugural major airline deployment, servicing all British Airways flights at JFK in partnership with dnata (Emirates Group subsidiary) as ground handler. HALO coordinates task, crew, vehicle, aircraft, and gate pairings.

Second US Hub Airport HALO is active at a second undisclosed US hub airport with Aerocharter as the ground handling partner.

Tokyo International Airport (Haneda) — Japan Airlines Announced July 2025, Moonware is testing HALO with Japan Airlines (JAL) and JAL Ground Service (JGS) at Tokyo Haneda. Focus areas: below-wing coordination, resource management, and real-time situational awareness. JAL and JGS expect reduced variability in turnaround times and streamlined ground service execution.

3.4 Moonware Master Plan — Bridge to Airfield Autonomy

Moonware's roadmap envisions handling aircraft autonomously from touchdown to takeoff across four phases:

Phase 1 — Software Foundation (Current): Deploy HALO with existing airlines, ground handlers, and airports to digitize and optimize manual ramp operations
Phase 2 — Autonomous Ramp Services: Leverage data and insights from Phase 1 to identify where autonomy augments ramp services; pilot autonomous GSE in defense (via NOVA) and emerging markets (Urban Air Mobility)
Phase 3 — Autonomous Ground Service Provider: Become a full autonomous ground handling operator; interface with ATC for autonomous surface movement of aircraft
Phase 4 — Advanced Air Mobility Integration: Airport infrastructure designed around service needs of diverse aircraft types including eVTOLs

Product Portfolio:

HALO: Commercial airside operating system
NOVA: Military Command-and-Control (C2) platform for flightline management, building on HALO's commercial foundation; tracks asset status, location, and mission phase; lays groundwork for Autonomous Aerospace Ground Equipment (A2GE)
ATLAS: Autonomous and electric pushback tug designed for eVTOL operations

Funding: $2.5M pre-seed, followed by $7M seed round to advance automated airfields.

4. GSE Fleet Coordination

4.1 Problem Structure

GSE fleet coordination is a multi-dimensional optimization problem combining:

Task allocation: Assigning ground handling tasks to heterogeneous GSE vehicles
Vehicle routing: Planning efficient paths across the apron with time windows
Scheduling: Sequencing vehicle trips within flight-dictated time constraints
Conflict resolution: Avoiding collisions on shared apron roadways and stand areas

The problem is formally a Vehicle Routing Problem with Time Windows (VRPTW) on airport surfaces. A single aircraft turnaround can involve 10–20 different pieces of equipment, and at a busy airport, hundreds of turnarounds occur simultaneously.

4.2 GSE Types Requiring Coordination

Nine canonical GSE types in scheduling optimization models:

Passenger stairs / Jet bridge — First on, last off
Baggage/cargo tractor and dollies — ULD transport
Belt loader — Bulk cargo loading/unloading
Fuel truck / hydrant dispenser — Refueling
Catering truck — Galley exchange
Potable water truck — Water replenishment
Lavatory service truck — Waste removal
Ground power unit (GPU) — Electrical power
Pushback tug — Aircraft movement from stand Additional: de-icing trucks (seasonal), air conditioning units (PCA), cleaning vehicles

4.3 Task Allocation Algorithms

Auction-Based Multi-Agent Allocation Recent research (2025) presents a centralized multi-agent task allocation and routing model for autonomous GSE. Ground handling tasks are modeled as single-vehicle pickup-and-delivery problems. The auction mechanism, inspired by temporal sequential single-item auctions, allocates tasks to the GSE vehicle that can service them at lowest cost (time + distance + energy). Each GSE vehicle "bids" on available tasks based on its current position, remaining capacity, and schedule constraints.

Integer Linear Programming (ILP) Classical formulation: minimize total service cost (travel time + waiting time + energy consumption) subject to:

Each flight's GSE requirements must be met within its service time window
Each GSE vehicle serves one task at a time
Travel times between stands are respected
Vehicle capacity and type constraints are enforced

Multi-Objective Optimization Bi-objective models for electric GSE scheduling jointly minimize:

Service time deviations (tasks completed within target windows)
Energy consumption and emissions
Fleet utilization balance

Solution algorithms include exhaustive methods, Clarke-Wright savings, column generation, genetic algorithms (GA), particle swarm optimization (PSO), and variable neighborhood search.

Multi-Agent Reinforcement Learning (MARL) MARL approaches treat each GSE vehicle as an autonomous agent learning to coordinate through shared or independent reward signals:

Multi-Agent Deep Deterministic Policy Gradient (MADDPG) handles simultaneous target assignment and path planning
Multi-Agent Proximal Policy Optimization (MAPPO) addresses shared autonomous vehicle scheduling
Agents learn collision avoidance, task prioritization, and cooperative behavior through simulated training environments

4.4 Multi-Vehicle Routing on Apron

Path Planning for Large Agents Prioritized Safe Interval Path Planning (SIPP) plans collision-free paths for large physical agents (GSE vehicles with significant footprint) on the apron. Unlike point-agent path planning, this accounts for vehicle dimensions, turning radii, and kinematic constraints.

Conflict-Free Scheduling

Temporal constraints: GSE must arrive at stand within service window (e.g., fuel truck after passengers deplane but before boarding)
Spatial constraints: Limited stand area means only certain GSE combinations can operate simultaneously
Precedence constraints: Some tasks must complete before others begin (e.g., deplaning before cleaning)
Safety zones: GSE must clear the aircraft safety perimeter before pushback

Digital Twin Approaches Multi-strategy cooperative scheduling based on digital twins creates a virtual replica of the apron environment, enabling real-time simulation of GSE movements, conflict detection, and schedule optimization before executing in the physical world.

4.5 Stand Sequencing

Stand sequencing determines the order in which aircraft are assigned to gates/stands, directly impacting GSE routing efficiency. Optimal stand assignment minimizes:

Total GSE travel distance between consecutive assignments
Towing requirements for remote stands
Passenger walking distance
Gate conflicts and buffer time requirements

5. Just-in-Time GSE Dispatch

5.1 Concept

Just-in-time (JIT) GSE dispatch aims to send each piece of ground support equipment to the stand precisely when needed — not too early (wasting equipment capacity and creating apron congestion) and not too late (delaying the turnaround and risking cascading delays). GSE scheduling inefficiencies are identified as the second most common cause of flight delays after air traffic control issues.

5.2 Predicting When Each GSE Type Is Needed

The prediction pipeline for JIT dispatch combines:

Flight-Level Predictions:

ELDT from ADS-B trajectory analysis triggers pre-positioning of arrival GSE
EIBT prediction (ELDT + estimated taxi-in time) provides the GSE dispatch target
Aircraft type determines specific GSE requirements (e.g., widebody requires high-loader vs. belt loader)
Load factor and cargo manifest predict unloading duration

Turnaround Phase Predictions:

Cascaded milestone prediction models estimate when each sub-process will complete
Completion of deplaning triggers cleaning and catering dispatch
Fuel order confirmation triggers fuel truck dispatch
Boarding completion prediction triggers pushback tug dispatch

Pattern-Based Scheduling:

Historical turnaround patterns by airline, aircraft type, time of day, and day of week
Seasonal adjustments (de-icing in winter, extended cooling in summer)
Ground handler staffing level correlation with turnaround duration

5.3 Reducing Idle Time and Apron Congestion

Telematics-Driven Fleet Management: Real-time GPS tracking of all GSE reveals usage patterns enabling:

Fleet right-sizing: Determine optimal number of each GSE type per terminal
Deadhead trip minimization: Route GSE directly between consecutive assignments
Predictive maintenance: IoT sensors detect anomalies and predict failures, reducing unplanned downtime by up to 40%
Charging optimization for electric GSE fleets: Schedule charging during predicted idle windows

Cooperative Scheduling Under Uncertainty: Flight schedules are inherently uncertain. Models that incorporate flight delay distributions and stochastic service times produce more robust GSE schedules. Cooperative scheduling algorithms adjust dynamically as actual flight times deviate from plan, reassigning GSE to maintain service levels.

Mixed Fleet Optimization: With the industry shift toward electric GSE (3,000+ electric units ordered globally in 2024-25), scheduling must account for:

Battery state-of-charge and range limitations
Charging station locations and availability
Mixed fuel/electric fleet operations during transition periods
Energy and emissions minimization alongside time objectives

GSE Pooling: Rather than each ground handler maintaining a dedicated fleet, pooled GSE models allow airport-wide sharing. Benefits include:

Economies of scale and fleet standardization
Simplified maintenance and training
Natural platform for introducing autonomous GSE under a single operating system
Reduced total fleet size through improved utilization

6. Integration with World Model

6.1 Turnaround Phase as Context for Prediction

For an airside autonomous vehicle (AV), the aircraft turnaround phase is critical contextual information that fundamentally changes the prediction landscape:

Phase-Dependent Scene Dynamics:

Arrival phase: High inbound traffic — marshalling vehicles, follow-me cars, arriving baggage trains approaching the stand. The AV must predict convergence patterns toward the assigned gate.
Unloading phase: Belt loaders and cargo tractors active on the stand apron; passenger buses or jet bridge in use. GSE movement is concentrated and relatively predictable.
Servicing phase: Maximum GSE density at stand — fuel truck, catering truck, water/lavatory vehicles, cleaning crews. The AV must navigate around a congested workspace with constrained sight lines.
Loading phase: Renewed cargo tractor activity; passenger flow resuming. Vehicles begin departing the stand area.
Departure phase: Rapid GSE withdrawal from safety zone, pushback tug positioning. High urgency movements with tight coordination requirements.

World Model Implications: The world model should encode turnaround phase as a discrete latent variable that modulates:

Expected agent density and types at each stand
Typical motion patterns (arrival vs. departure flow directions)
Urgency and right-of-way expectations (pushback tug has priority)
No-go zones that change dynamically (safety perimeter enforcement)

6.2 Flight Schedule Priors

Flight schedules provide strong priors for the world model's prediction of future scene states:

Structured Temporal Priors:

AODB flight schedule gives exact expected arrival/departure times for every gate
Known aircraft type determines expected GSE ensemble and turnaround duration
Schedule patterns are highly regular (same flights at same gates at same times daily)
Delay propagation models predict how upstream delays cascade to local operations

Multi-Scale Temporal Reasoning:

Minutes-scale: Which turnaround phase is active at each gate? What GSE movements are imminent?
Tens-of-minutes-scale: When will the next aircraft arrive at an adjacent stand? How will apron traffic patterns shift?
Hours-scale: What is the overall traffic density trend? Are we approaching a departure bank or arrival wave?

Conditional Generation: The world model can condition its future scene predictions on flight schedule information:

P(scene_t+k | scene_t, flight_schedule, turnaround_phases)

This is substantially more informative than unconditional prediction. A world model that knows Flight BA117 is scheduled to push back from Gate B22 in 8 minutes can anticipate pushback tug approach, GSE withdrawal, and jet bridge retraction — even before observing these events.

6.3 Temporal Reasoning Architecture

Hierarchical Temporal Model: The world model should maintain multiple temporal representations:

Immediate horizon (0–10s): Physics-based motion prediction for all visible agents (vehicles, pedestrians, GSE). Standard BEV occupancy prediction and trajectory forecasting.
Near horizon (10s–5min): Activity-conditioned prediction. Given the current turnaround phase, predict likely GSE arrivals/departures. Attention mechanisms over the turnaround activity graph.
Planning horizon (5–60min): Schedule-conditioned prediction. Given the flight schedule and current state, predict the evolution of the operational scene. This enables proactive route planning that avoids future congestion.

Graph-Based Turnaround Representation: Model the turnaround as a temporal activity graph where:

Nodes represent turnaround sub-processes (deplaning, fueling, catering, etc.)
Edges represent precedence and parallelism relationships
Node states encode progress (not started / in progress / completed)
Edge weights represent predicted transition times
The graph evolves over time as milestones are reached

This turnaround graph can be encoded via a Graph Neural Network (GNN) and fused with the world model's spatial scene representation, providing structured temporal context that raw sensor data alone cannot capture.

Delay Propagation Awareness: Spatiotemporal graph models (GAT-LSTM, STPN) from the aviation domain demonstrate how delays propagate across the airport network. The world model can incorporate these patterns to predict how a delayed arrival at one gate will cascade to affect adjacent gates, taxiways, and apron traffic patterns over the next hour.

7. Communication Infrastructure and Fleet Coordination

7.1 V2X for GSE Fleet

Vehicle-to-Everything (V2X) communication enables GSE vehicles and autonomous AVs to share position, speed, intent, and status with each other and with infrastructure. Two competing standards exist:

DSRC (Dedicated Short-Range Communications)

IEEE 802.11p-based, operating at 5.9 GHz
Range: 300–1,000 meters
Latency: <10ms
Designed for ad-hoc, distributed operation
Well-suited for direct vehicle-to-vehicle (V2V) safety messages on the apron
No cellular infrastructure dependency

C-V2X (Cellular V2X)

3GPP standard (LTE-V2X in Release 14, 5G NR-V2X in Release 16)
Two interfaces: PC5 (direct short-range) and Uu (cellular network)
PC5 provides DSRC-comparable latency for safety-critical V2V messages
Uu interface leverages existing cellular infrastructure for V2N (vehicle-to-network) communication
5G NR C-V2X offers ultra-low latency (<5ms), high reliability (99.999%), and greater capacity

Airport-Specific V2X Considerations:

Apron environment is geographically bounded — infrastructure can be purpose-built
Mixed traffic: autonomous vehicles, human-driven GSE, pedestrians, aircraft
Safety criticality: runway incursion prevention, pushback coordination
Dense vehicle operations at stands require reliable short-range communication
C-V2X PC5 mode is advantageous for direct GSE-to-GSE coordination without network dependency
Hybrid DSRC/C-V2X architectures proposed for comprehensive coverage

7.2 5G/CBRS Airport Deployments

Citizens Broadband Radio Service (CBRS) in the 3.5 GHz band enables airports to deploy private 5G networks without licensed spectrum costs, providing dedicated, secure, low-latency connectivity for airside operations.

Dallas Fort Worth International Airport (DFW)

200+ access points deployed with private 5G backbone
Supports asset tracking, autonomous vehicle trials, and digital twins
Autonomous tugs, computer vision, and AI-driven maintenance deployed
Completed three comprehensive proofs of concept in ramp, cargo, and terminal services before full buildout

Purdue University Airport (Ericsson Partnership)

Private 5G network over CBRS deployed in collaboration with Ericsson and Saab
Supports flight coordination, real-time security, drone detection, and autonomous ground equipment
Reported up to 30% productivity gains with measurable safety improvements
Serves as innovation hub for aviation technology testing

Singapore Changi Airport (Nokia Partnership)

Nokia and Changi Airport Group partnered with M1 Limited for private 5G network (rolled out 2024)
Supports autonomous ground vehicles, baggage handling systems, drone-based maintenance inspections
Edge computing for predictive maintenance and real-time asset tracking
Digital twin operations and intelligent maintenance
First fleet of fully driverless autonomous tractors deployed following 5,000+ test trips
Two autonomous tractors operational for baggage transfer between T1 and T4; fleet expanding to 24 by 2027

Amsterdam Schiphol

Private 5G pilot supporting IoT-enabled predictive maintenance, smart baggage handling, and autonomous ground vehicles

Los Angeles (LAX) — Lufthansa Cargo

Replaced Wi-Fi with private 5G to accelerate warehouse operations

CBRS Technical Characteristics for Airport Use:

Supports connected assets, integrated real-time communication, digital load control
Remote uploading/offloading of data
Secure, dedicated spectrum avoiding interference from public cellular
Sufficient bandwidth for video analytics, real-time tracking, and autonomous vehicle telemetry

7.3 Coordination Protocols

Fleet Management System Architecture: A centralized fleet management system for airside autonomous vehicles typically includes:

Central Coordinator: Receives tasks from AODB/ground handler dispatch systems, runs optimization algorithms, and issues assignments to vehicles
Vehicle Agents: Each autonomous vehicle runs local perception, planning, and control; communicates position, status, and task progress to coordinator
Infrastructure Layer: Smart intersections, stand occupancy sensors, and ADS-B receivers provide situational awareness
Communication Bus: 5G/CBRS for reliable wide-area coverage; C-V2X PC5 for direct V2V safety messages; Wi-Fi 6E as backup

Coordination Message Types:

Task Assignment: Central coordinator dispatches vehicle to specific stand for specific service
Position Broadcast: Vehicles broadcast position, heading, speed at 10 Hz for collision avoidance
Intent Sharing: Vehicles share planned trajectories for cooperative path planning
Status Updates: Task progress (en route, arrived, servicing, complete) for schedule tracking
Emergency Stop: Safety-critical broadcast for immediate halt of all vehicles in vicinity

7.4 Moonware, EVIE, and reference airside AV stack Integration

Moonware HALO as Orchestration Layer: HALO's centralized coordination engine is architecturally positioned to serve as the dispatch and orchestration layer for autonomous GSE fleets. Its real-time task allocation, schedule integration, and timestamped activity tracking provide the command interface that autonomous vehicles require. Moonware's Phase 2 roadmap explicitly targets autonomous ramp services, leveraging HALO data to inform where autonomy augments operations.

reference airside AV stack autonomous baggage/cargo tug:

All-electric autonomous baggage/cargo tractor combining tractor and dolly functions
Unique sideways drive system enables rotation on the spot and lateral movement
Bi-directional robotic arms for autonomous ULD loading/unloading
Tested at six global airports including Changi, Schiphol, and CVG (Cincinnati)
Partnership with Aviation Solutions for global commercialization
IAG program commenced at Cincinnati/Northern Kentucky International Airport
LiDAR, radar, and camera sensor suite for airside navigation
Integration requires fleet management API for task receipt and status reporting

EVIE Autonomous:

Electric modular chassis platform with multiple pod configurations (Airside Pod, Cargo Pod, Shuttle)
LiDAR providing 360-degree view, real-time radar for velocity estimation, high-resolution cameras with AI object classification
Designed to integrate with existing airport GSE, enabling retrofit of self-driving capability
Cargo Pod handles aircraft brakes (baggage containers) and ULDs on a single platform
Addresses airport staffing challenges with reliable autonomous airside operations

Integration Architecture for Autonomous GSE Fleet:

┌─────────────────────────────────────────────────────┐
│              AODB / Flight Schedule                  │
│        (Flight times, gate assignments, A-CDM)       │
└──────────────┬──────────────────────────┬───────────┘
               │                          │
               v                          v
┌──────────────────────┐   ┌──────────────────────────┐
│   Moonware HALO      │   │  Turnaround Prediction   │
│  Ground Traffic       │   │  (Cascaded GBRT, GNN,    │
│  Control Platform     │   │   Computer Vision)       │
│  - Task allocation    │◄──│  - Phase detection       │
│  - Crew dispatch      │   │  - TOBT prediction       │
│  - Schedule mgmt      │   │  - GSE timing forecast   │
└──────────┬───────────┘   └──────────────────────────┘
           │
           v
┌──────────────────────────────────────────────┐
│        Fleet Management / Dispatch            │
│  - JIT GSE assignment                         │
│  - Multi-vehicle routing (VRPTW)             │
│  - Conflict-free scheduling                   │
│  - Dynamic re-planning on delay              │
└──────┬──────────┬──────────┬─────────────────┘
       │          │          │
       v          v          v
┌──────────┐ ┌──────────┐ ┌──────────┐
│ reference airside AV stack  │ │  EVIE    │ │ Legacy   │
│ Auto-    │ │ Airside  │ │ Human-   │
│ DollyTug │ │ Pod      │ │ Driven   │
│          │ │          │ │ GSE      │
└──────────┘ └──────────┘ └──────────┘
       │          │          │
       └──────────┴──────────┘
                  │
           5G / C-V2X / CBRS
        (Position, status, intent)

Key Integration Requirements:

Standardized Task API: Common interface for receiving dispatch commands (go to stand X, perform service Y, report completion)
Position and Status Telemetry: Vehicles report location and task state via 5G/C-V2X at sufficient frequency for fleet-level optimization
Safety Interlock: Infrastructure-level safety system that can command emergency stops, enforce no-go zones, and manage aircraft safety perimeters
Mixed Fleet Support: Orchestration layer must handle both autonomous and human-driven vehicles during transition period
A-CDM Feedback Loop: Actual GSE arrival/service/departure times feed back into turnaround prediction models, improving future forecasts

Summary and Implications for Airside AV World Models

Aircraft turnaround prediction and GSE fleet coordination represent a rich domain of structured temporal context that airside autonomous vehicles can leverage. The key insight is that airport operations are far more predictable than general road traffic — flights follow published schedules, turnarounds follow known sequences, and GSE movements are task-driven rather than free-form.

For the world model specifically:

Turnaround phase should be encoded as a discrete context variable modulating scene predictions
Flight schedules from AODB provide strong temporal priors over minutes-to-hours horizons
A-CDM milestone data enables structured prediction of when specific GSE types will appear at specific locations
Graph neural networks can model both the spatial structure of the apron and the temporal structure of turnaround operations
Computer vision systems (Assaia ApronAI) provide ground-truth turnaround phase labels for training

For fleet coordination:

JIT GSE dispatch requires tight integration between turnaround prediction and vehicle routing
Multi-agent path planning with temporal constraints (VRPTW) is the core optimization problem
Private 5G/CBRS networks provide the communication backbone for real-time fleet coordination
Moonware HALO provides a proven orchestration architecture that bridges the gap between flight operations and vehicle dispatch
The transition from human-driven to autonomous GSE (reference airside AV stack, EVIE) requires a software layer that can manage mixed fleets

The convergence of turnaround prediction ML, fleet coordination algorithms, private 5G infrastructure, and autonomous GSE platforms creates a complete stack for intelligent airside operations — one where the autonomous vehicle's world model is informed not just by what it sees, but by the structured operational context of the airport it operates within.

SLAM Methods

Methods

Aircraft Turnaround Prediction, GSE Fleet Coordination, and Operational Integration for Airside Autonomous Vehicles ​

1. Aircraft Turnaround Phases ​

1.1 Standard Turnaround Sequence ​

1.2 Typical Durations by Aircraft Type ​

1.3 IATA AHM Ground Handling Standards ​

2. Turnaround Prediction ​

2.1 Predicting Phase Transitions ​

2.2 ML Approaches ​

2.3 Data Sources ​

3. Moonware HALO Platform ​

3.1 Overview and Architecture ​

3.2 Performance Claims ​

3.3 Deployments ​

3.4 Moonware Master Plan — Bridge to Airfield Autonomy ​

4. GSE Fleet Coordination ​

4.1 Problem Structure ​

4.2 GSE Types Requiring Coordination ​

4.3 Task Allocation Algorithms ​

4.4 Multi-Vehicle Routing on Apron ​

4.5 Stand Sequencing ​

5. Just-in-Time GSE Dispatch ​

5.1 Concept ​

5.2 Predicting When Each GSE Type Is Needed ​

5.3 Reducing Idle Time and Apron Congestion ​

6. Integration with World Model ​

6.1 Turnaround Phase as Context for Prediction ​

6.2 Flight Schedule Priors ​

6.3 Temporal Reasoning Architecture ​

7. Communication Infrastructure and Fleet Coordination ​