PaperlayerView exact revision on arXiv

Source-linked AI summary

Demand-Driven Vertiport Siting and Discrete-Event Fleet Simulation for On-Demand Urban Air Mobility Network Design

Hossein Z. Saghazadeh, Yonas Ayalew, Reza Ahmari, Parham Kebria, Abdollah Homaifar

arXiv:2608.14974v1cs.AIeess.SY

TL;DR

UAM network design lacks integrated evidence linking demand-based vertiport siting with event-driven fleet operations and operationally grounded travel-time savings. This paper combines clustering-based siting, discrete-event simulation, and door-to-door analysis, finding that higher demand favors larger station–fleet networks while deadhead flights persist.

  • Problem

    Existing methods often lack event-driven evaluation of vertiport locations and operationally grounded estimates of non-flight travel-time penalties.

  • Method

    The framework constructs spatial demand, clusters candidate vertiports under range and spacing constraints, and evaluates networks with discrete-event simulation linked to an eVTOL performance model.

  • Results

    The preferred Greater Los Angeles design grows from four stations and four vehicles at low demand to sixteen stations and twelve vehicles at the highest tested demand, while deadhead flights persist.

  • Takeaways & Limitations

    Persistent deadhead flights indicate that spatial origin–destination imbalance remains an operational burden despite larger fleets.

  • Takeaways & Limitations

    The case study uses Monte Carlo origin–destination generation as an operational stress test rather than a calibrated empirical demand model.

Abstract

from arXiv · show

This paper presents a demand-driven framework for on-demand Urban Air Mobility (UAM) network design that links vertiport siting, fleet simulation, and door-to-door travel-time feasibility. Demand is estimated from commuter and passenger activity data, converted into spatial trip-end points, and clustered using K-means to generate candidate vertiport locations. Candidate networks are screened using range and minimum station-spacing constraints, then evaluated with a discrete-event simulation that models multi-vehicle dispatch, deadhead relocation, battery swaps, and service regularity. Flight time and energy consumption are computed using a point-mass eVTOL performance model. In a Greater Los Angeles case study, the preferred design expands from four stations and four eVTOLs at low demand to sixteen stations and twelve eVTOLs at the highest tested demand level. Results show that larger fleets improve completion time and vehicle-arrival regularity but do not eliminate deadhead flights, indicating that spatial demand imbalance remains an operational burden. The travel-time savings analysis further suggests that UAM is most defensible for longer or congestion-heavy trips where sufficient non-flight time remains after accounting for flight time.

I. INTRODUCTION · II. PROBLEM DEFINITION AND EVALUATION FRAMEWORK

The paper develops a demand-driven UAM network-design pipeline that connects demand construction, constrained vertiport siting, discrete-event fleet operations, and door-to-door travel-time feasibility. It addresses gaps in event-driven evaluation and simulation-based non-flight travel-time assessment through a formal station-and-fleet framework.

  • I. INTRODUCTION: UAM design must coordinate vertiport siting and fleet planning as deployments progress from low-tempo services toward denser networks.Prior work indicates that scalable UAM depends on more than aircraft performance, including vertiport availability and operational planning.
  • I. INTRODUCTION: Existing studies cover demand estimation, door-to-door trip benefits, clustering-based siting, optimization-based placement, and operational policies.Clustering offers scalable, interpretable candidates, while clustering choices can affect predicted benefits and equity outcomes.
  • I. INTRODUCTION: Two gaps motivate the framework: siting methods often lack event-driven simulation of simultaneous operations and deadhead repositioning, while savings analyses assume non-flight penalties.The paper instead links non-flight penalties to operational metrics obtained from fleet simulation.
  • I. INTRODUCTION: The proposed pipeline constructs spatial demand, generates clustered candidate networks under range and spacing constraints, and evaluates designs with multi-vehicle dispatch and deadhead relocation.The pipeline also couples simulation with an eVTOL performance model.
  • II. PROBLEM DEFINITION AND EVALUATION FRAMEWORK: The system represents N demand trip-ends in X, K vertiports in V, and F identical eVTOLs providing station-to-station flights.Each passenger trip contributes an origin and destination point within the geographic study region.
  • II. PROBLEM DEFINITION AND EVALUATION FRAMEWORK: Each trip is modeled door-to-door as ground access, an inter-station flight, and ground egress, with nearest-station assignment inducing a station-level OD matrix Q.The origin and destination are routed through their assigned nearest stations, and q_ij counts requested passengers from station i to station j over the evaluation horizon.
  • II. PROBLEM DEFINITION AND EVALUATION FRAMEWORK: Candidate designs select K, V, and F while satisfying Δ_min ≤ d(v_i,v_j) ≤ Δ_max for all i ≠ j.Δ_max enforces feasible inter-station missions, whereas Δ_min prevents redundant co-location and supports practical spacing.
  • II. PROBLEM DEFINITION AND EVALUATION FRAMEWORK: The framework omits airspace conflicts, vertiport pad-capacity limits, and detailed passenger arrival queuing, while assuming stations can charge or swap batteries as needed.Travel time decomposes into flight and non-flight components, including access, processing, station-level MVAI, and egress, supporting savings-feasibility analysis.

III. METHODOLOGY · A. Demand Construction and Station Siting

The methodology constructs potential UAM demand from commuter, passenger, and early-adopter characteristics, expands aggregated trip ends into spatial demand points, and clusters them to generate candidate vertiport locations. Candidate station counts are compared using silhouette quality and retained subject to inter-station distance constraints.

  • A. Demand Construction and Station Siting: Historical UAM usage data are unavailable, so potential demand is estimated from trip length, preferred mode, and income characteristics.LODES home–work commuter flows provide the primary commuter-trip basis.
  • A. Demand Construction and Station Siting: ACS commuter totals calibrate LODES flows to reduce bias, while airport and inter-city rail passengers add a separate demand component.APT demand is allocated across ZCTAs in proportion to population.
  • A. Demand Construction and Station Siting: The eligibility filter targets car users with annual income above $75k as one plausible near-term early-adopter demand scenario.The paper explicitly distinguishes this filter from an equity objective or recommendation to prioritize higher-income areas.
  • A. Demand Construction and Station Siting: ZCTA-level potential demand combines outgoing and incoming trip ends, with incoming demand defined by destination ZCTA under the same criteria.This produces demand at the ZCTA level before spatial expansion for station siting.
  • A. Demand Construction and Station Siting: Demand points are expanded from ZCTA representative locations by placing one point when Dz ≥1 and adding annularly jittered points accepted within the ZCTA polygon.The expansion avoids multiple demand points accumulating at a single representative location.
  • A. Demand Construction and Station Siting: K-means clusters the expanded demand-point set X, using cluster centroids as candidate vertiport locations for a selected station count K.The method is presented as computationally efficient and interpretable for large-scale demand sets.
  • A. Demand Construction and Station Siting: Average silhouette values compare candidate station counts, with each silhouette value bounded within [−1,1].The coefficient contrasts mean within-cluster distance with the minimum mean distance to another cluster.

B. Discrete-Event On-Demand Operations Model

The discrete-event simulator evaluates candidate station–fleet designs by dispatching available eVTOLs to passenger trips or deadhead relocations while tracking remaining demand and vehicle availability. Each passenger-carrying or deadhead leg is simulated with a point-mass eVTOL model using an energy-minimizing cruise altitude and computed flight time and energy.

  • Event logic: dispatch and deadhead relocation: At each availability event, the simulator tracks remaining OD trips and vehicle locations, then dispatches the next-available eVTOL to a feasible passenger request or relocation.Passenger trips are selected from feasible OD requests, prioritizing the largest remaining OD entry and breaking ties by nearest destination.
  • Monte Carlo demand and initial conditions: Monte Carlo runs redistribute OD flows stochastically while preserving a fixed total passenger budget, testing how configurations respond to demand imbalance rather than forecasting calibrated empirical demand.The randomized runs evaluate each candidate under multiple route-demand realizations with the same total demand level.
  • Event logic: dispatch and deadhead relocation: Deadhead flights send idle eVTOLs to the nearest admissible station without a currently present vehicle and unmet demand, helping them re-enter service.Relocation destinations are restricted to S_dem \ S_occ, with the nearest station selected by distance.
  • Per-leg flight simulation and minimum-energy altitude: Each passenger-carrying or deadhead leg follows five flight phases in a point-mass eVTOL model, with cruise altitude selected to minimize energy and flight time and energy computed for the leg.The model is implemented in MATLAB/Simulink (R2025) and uses velocity-pursuit guidance to regulate heading toward the destination.

Z TPVP

This section defines the discrete-event simulation metrics used to evaluate completion time, vehicle-arrival regularity, battery use, and deadhead burden. These metrics, together with station and fleet counts, support preferred network selection under the stated assumptions.

  • Service regularity: MVAI measures station-level vehicle-arrival regularity from time-ordered vehicle-arrival timestamps over the simulation horizon.Lower MVAI indicates more frequent vehicle availability, but it is not a passenger waiting-time estimate because passenger arrivals and queues are not modeled.
  • Completion time: Evacuation/completion time combines the fleet’s maximum cumulative flight time with fixed processing times for battery changes, passenger service, flown legs, and airspace clearance.The processing terms include battery-change duration, boarding and deboarding times, and clearance time for each flight.
  • Energy use: Battery swaps are computed from total network energy consumption across all flown legs and the fixed available energy per battery.Network energy is accumulated over the set of flown legs before determining battery-change events.
  • Relocation burden: Deadhead burden is summarized by the deadhead ratio based on the number of deadhead legs executed during the simulation.Deadhead legs are included in the total flown-leg count used by the completion-time formulation.
  • Design selection: ET, MVAI, DHR, CB, NS, and NE are passed to the design-selection criterion to identify preferred network configurations.NS denotes Number of Stations and NE denotes Number of eVTOLs.

C. Design Selection Criterion

Candidate network designs are evaluated using aggregated operational metrics and a weighted cost function. The adopted weights prioritize completion time and service regularity, followed by energy-related and other operational or infrastructure penalties.

  • Metric aggregation: Each candidate design (K,V,F) is evaluated with MVAI, ET, CB, and DHR, with Monte Carlo metrics aggregated into one estimate per design when enabled.The estimates are denoted generically by ET(K,V,F), MVAI(K,V,F), DHR(K,V,F), and ¯NC….
  • Weighted cost function: The weights satisfy wET + wMVAI + wDHR + wCB + wNE + wNS = 1 and wi > 0.The cost function combines these weighted normalized terms for candidate-design comparison.
  • Planning preference: wET = wMVAI = 2wCB = 4wDHR = 4wNE = 4wNS prioritizes temporal metrics, followed by the energy-related term and remaining cost-related terms.These weights encode the planning preference used in the study rather than a universal economic cost model.

D. Door-to-Door Travel-Time Savings Feasibility

The section evaluates UAM travel-time savings using a door-to-door decomposition into flight and non-flight components. A savings target is feasible only when non-flight time fits within the allowable slack relative to modeled ground travel.

  • Door-to-Door Formulation: Door-to-door travel time is expressed as the sum of flight and non-flight components, with R denoting station-to-station separation.The separation may be represented as R = d(v_i,v_j) for i ≠ j.
  • Ground-Time Baseline: Ground travel time uses effective average speed v, circuity factor κ = 1.42, and access/egress radius δ ≤5km.The model is defined as T_ground(R;v) = κ R+δ.
  • Savings Feasibility: A savings target s ∈(0,1) is achievable only if T_nonflight(R) ≤τ(R;v,s), where τ(R;v,s) = (1 −s)T_ground(R;v) −T_flight(R).The condition follows from requiring T_door(R) ≤(1− s)T_ground(R;v); positive slack τ > 0 indicates an available non-flight-time budget.

IV. CASE STUDY SETUP

The Greater Los Angeles case study applies the demand–siting–simulation pipeline to approximately 2.96 × 10^5 baseline trip-ends and evaluates feasible K-means station networks with varying fleet sizes. Feasibility limits the tested network to 16 stations, with designs assessed across K = {4,8,12,16} and fleet sizes F = {2 : 2 : K}.

  • Case-study region and demand: The case study applies the demand–siting–simulation pipeline to Greater Los Angeles using commuter, airport, and inter-city rail passenger data.Spatial computations use the UTM projected CRS with the Euclidean metric, while WGS84 coordinates are used only for visualization.
  • Case-study region and demand: 2.96 × 10^5 trip-ends form the approximate candidate passenger pool and baseline demand for defining scenarios.The pool is produced under the adopted early-user eligibility filter.
  • Candidate network design: K = {4,8,12,16} feasible station sets are evaluated with multiple fleet sizes F = {2 : 2 : K} to quantify station–fleet tradeoffs.Candidate locations are generated by the K-means siting procedure and screened under the feasibility constraints.
  • Candidate network design: 16 stations cap the tested network because adding stations produces infeasible missions and a lower silhouette value.The cap follows screening with Δmin ≤ d(v_i,v_j) ≤ Δmax for all distinct station pairs.

V. RESULTS

The results report a Greater Los Angeles case study in which candidate designs are evaluated through discrete-event simulation, and preferred designs are selected by minimizing J. They examine how (K∗,F∗) scales with demand and summarize operational tradeoffs across six metrics.

  • The Greater Los Angeles case study evaluates candidate designs (K,V,F) using the discrete-event simulation described in Section III-B.
  • Preferred designs for each demand scenario are selected by minimizing J in Equation (16).
  • The results examine how (K∗,F∗) scales with demand.
  • Operational tradeoffs are summarized using ET, MVAI, DHR, CB, NE, and NS.

A. Preferred Candidate Designs Across Demand Scenarios

The preferred design expands in station–fleet scale as demand rises, from (4,4) at D = 100 to (8,8) at D = 500–1000 and the maximum feasible station count at higher demand. At D = 2000, fleet-abundant tests show improved temporal performance with more vehicles, but battery swaps increase and deadhead flights remain necessary.

  • Preferred designs: (4,4) at D = 100 expands to (8,8) at D = 500–1000, then reaches the maximum feasible station count at higher demand.The preferred design is reported as (K∗,F∗), with station locations determined by the corresponding feasible K-means station set.
  • Fleet-abundant sensitivity: At D = 2000 with F = 4K, increasing fleet size substantially decreases ET and MVAI, showing that temporal performance is fleet-limited.The fleet-abundant case evaluates several station counts while scaling vehicles as F = 4K.
  • Fleet-abundant sensitivity: Higher fleet levels increase CB while DHR remains nonzero, so battery-swap activity rises without eliminating deadhead flights.Deadhead flights continue moving vehicles toward remaining demand and supporting network operation.
  • Fleet-abundant sensitivity: The D = 2000 fleet-abundant case isolates fleet availability rather than representing a cost-optimal deployment, and spatially imbalanced OD demand preserves repositioning needs.Adding vehicles alone cannot remove repositioning requirements when demand is spatially imbalanced.

B. Station–Fleet Tradeoffs and Objective Landscape · C. Door-to-Door Travel-Time Savings Feasibility

The objective’s low-cost station–fleet region shifts toward larger designs as demand increases, while larger fleets improve completion time and service regularity. Door-to-door feasibility depends on allowable non-flight time, making UAM more defensible for longer or congestion-heavy trips.

  • B. Station–Fleet Tradeoffs and Objective Landscape: As demand rises from D ∈{100,500,1000,2000}, the low-cost region of J shifts toward larger station–fleet designs, with selected designs inside these basins.Nearby tested (K,F) choices therefore have limited cost sensitivity.
  • B. Station–Fleet Tradeoffs and Objective Landscape: For D = 2000, increasing fleet size F at fixed station count K reduces completion/evacuation time ET and decreases mean vehicle arrival interval MVAI.The same fleet increase also raises realized flight frequency, indicating more frequent service.
  • B. Station–Fleet Tradeoffs and Objective Landscape: For D = 2000, larger F improves service regularity while increasing realized flight frequency across tested station configurations.These outcomes are reported as discrete-event simulation results over station counts K and fleet sizes F.
  • C. Door-to-Door Travel-Time Savings Feasibility: At v = 80 km/h, the maximum allowable non-flight time budget decreases as the travel-time savings target increases and generally grows with station-to-station range.The relationship is shown for multiple savings targets and ranges.
  • C. Door-to-Door Travel-Time Savings Feasibility: UAM is more defensible for longer or congestion-heavy trips when sufficient non-flight time remains after accounting for flight time.The travel-time feasibility analysis links this condition to the allowable non-flight budget.
  • C. Door-to-Door Travel-Time Savings Feasibility: A Santa Clarita–LAX station pair spans approximately 57 km with an eVTOL flight time of about 14 min, while Google Maps reported 65–120 min of ground travel.For a 40% savings target, the resulting allowable non-flight budget is approximately [20,52] min after subtracting flight and 5 min access time.

VI. CONCLUSION AND FUTURE WORK

The paper presents a demand-driven UAM network-design framework linking demand construction, vertiport siting, discrete-event operational evaluation, and door-to-door travel-time analysis. The case study shows that demand-responsive scaling improves operations but does not eliminate deadhead flights, while identified modeling limitations motivate expanded empirical, infrastructure, scheduling, and airspace analyses.

  • Contributions: The framework links demand construction, K-means vertiport siting under range and spacing constraints, discrete-event operational evaluation, and door-to-door travel-time savings analysis.Demand was estimated from public commuter and passenger activity data.
  • Case-study findings: Four stations and four vehicles at low demand expand to sixteen stations and twelve vehicles at the highest tested demand level.The preferred station–fleet configuration increases with demand in the Greater Los Angeles case study.
  • Case-study findings: Increasing fleet size reduced completion time and improved vehicle-arrival regularity but did not eliminate deadhead flights driven by fleet availability and spatial OD imbalance.Empty repositioning therefore remains an operational burden despite larger fleets.
  • Limitations: The Monte Carlo OD generation is an operational stress test rather than a calibrated empirical OD model, and K-means is a scalable siting baseline rather than a globally optimal method.The DES also omits passenger arrivals, queues, vertiport capacity, alternative relocation policies, weight sensitivity, and airspace conflicts.
  • Future work: Future work will perturb the empirical OD matrix and extend siting and operations models with facility-location constraints, queueing, capacity, routing, dispatch policies, weight sensitivity, and airspace conflicts.Proposed constraints include demographic, land-use, equity, and regulatory factors.
Loading 2608.14974v1…