Paper
UAV Route Planning and Augmented Reality
Using Detroit Metropolitan Airport as a testbed: mapping airspace and regulatory constraints, designing four eVTOL arrival routes that stay clear of airline traffic, and building an open-source planning and 3D replay tool. Accepted to ICAS 2026.
A research project with Joby and Detroit Metropolitan Airport (DTW). The paper was accepted to ICAS 2026.
The research asks how electric vertical takeoff and landing (eVTOL) aircraft could fly as air taxis between a major city and a hub airport. According to the paper’s review, no such route has operated yet. eVTOLs could cut city-to-airport travel time by up to 90%, but the airspace above a hub is already full of airliners on fixed approach corridors, so a new aircraft needs approach routes, holding positions and a landing site that stay out of their way.
Using DTW as the testbed, the paper designs four candidate routes from Ann Arbor and downtown Detroit to a pad beside DTW’s Concourse C. Each takes about 12 to 20 minutes and stays within the Joby S4 performance envelope. My part covered the airspace and regulatory analysis, route design, the kinematic model and the web planning tool.
Why DTW
DTW already has an air mobility corridor between Detroit and Ann Arbor that supports beyond-visual-line-of-sight (BVLOS) operations. It has an underused concourse that could become a vertiport, spare apron space, and support from state and local government for advanced air mobility trials. It is a real hub with room to test without disrupting scheduled flights, which makes it a good place to study the problem before moving to more crowded airports such as New York’s JFK.
Airspace and regulations
Before designing routes, I mapped out where near the airport an eVTOL cannot fly or faces restrictions.
DTW is surrounded by Class B airspace, shaped like an upside-down layered cake whose floor gets lower closer to the airport. The FAA Aeronautical Information Manual restricts unauthorized low-altitude entry near hubs. The two innermost shelves matter most for eVTOLs: a surface core out to 8 NM reaching 10,000 ft, and a first shelf out to 10 NM with its floor at 2,500 ft.
Airliner instrument approaches are designed under the U.S. Standard for Terminal Instrument Procedures (TERPS) and come with obstacle clearance surfaces that eVTOL routes must avoid. FAA Order JO 7110.65 adds horizontal and vertical radar and wake separation minimums, and requires holding areas to absorb swings in traffic.
NASA’s Urban Air Mobility Concept of Operations describes interstitial airspace, the little-used space beneath the Class B floors and between streams of instrument traffic. That is where eVTOLs are most likely to get in and out.
The approach corridors were rebuilt for the precision runway monitor (PRM) instrument approach, which takes up the most airspace, so the conclusions lean conservative. With historical tracks from NASA’s Sherlock data warehouse overlaid, airliners stay almost entirely in the corridors and on the upper Class B shelves, and the low-altitude band beneath them is largely unused. The eVTOL routes are placed below the airline traffic for that reason.
Route design
The scenario is DTW’s south-flow configuration, using Runways 22R, 22L and 21R. I designed the routes by hand against three principles:
- Keep a lateral and vertical safety buffer from the PRM approach volumes
- Follow existing ground transport corridors such as railroad lines where possible, which reduces noise over neighbourhoods and makes use of existing BVLOS infrastructure
- Keep approach altitudes below or between the approach corridors, and where possible below the traffic pattern altitude of nearby airports such as Willow Run
The Ann Arbor route comes in from the west, laterally offset from the Runway 22R approach corridor. The Detroit route comes in from the northeast through low-altitude sectors that current arrivals seldom use.
Each route has a holding fix before final approach. The fix has to sit over largely unpopulated green space, so the aircraft can land nearby if a hold outlasts its energy reserve, and close enough to the airport for controllers to keep it in sight.
Each route was tested with two holding strategies:
| North hold | Lateral hold (west / east) | |
|---|---|---|
| Position | North of the field, behind the main arrival and departure flow | Abeam the runway, laterally clear of the flow |
| Wake turbulence | More exposure, often behind heavy jets | Less exposure |
| Controller workload | Crosses only an inactive crosswind runway, so a pre-arranged letter of agreement (LOA) could replace individual clearances | Crosses an active runway, needing a specific clearance and a gap in arrivals and departures |
The lateral hold has less wake risk but crosses an active runway, which adds controller workload, one of the main concerns about eVTOLs at Class B airports. The north hold trades the other way. The paper does not name a single best option, and the choice depends on which cost the airport cares about more.
| Origin | Hold | Waypoints | Flight time | Max altitude (AGL) |
|---|---|---|---|---|
| Ann Arbor (MCity) | North | 13 | 20 min 24 s | 1,154 ft |
| Ann Arbor (MCity) | West | 14 | 16 min 36 s | 1,154 ft |
| Detroit (Michigan Central) | North | 8 | 11 min 55 s | 1,654 ft |
| Detroit (Michigan Central) | East | 14 | 15 min 32 s | 1,654 ft |
Kinematic model
To check that the routes can actually be flown, I built a point-mass kinematic model that smooths discrete waypoints into time-resolved trajectories. Six-degree-of-freedom simulations such as NASA’s GUAM are more accurate but too slow to check a route while waypoints are being dragged. A point-mass model covers only turning, climbing and braking, which is enough to judge feasibility and fast enough to run behind a web page.
The input is an ordered list of waypoints, each with latitude, longitude, height above ground and a hold flag. The output gives 3D position, groundspeed, heading, vertical rate, bank angle and load factor at each time step. Flight is split into five phases: takeoff, forward flight, hold, landing and taxi. In forward flight, speed follows a stopping-distance rule so the aircraft slows down in time for the next waypoint.
Performance limits come from the Joby S4: 90 kt cruise, 492 ft/min climb, 0.15 g maximum acceleration, 0.10 g service deceleration and a turn rate of 3° per second. With these numbers the minimum turn radius is about 884 m. Waypoints that are too close or joined by too sharp a turn make a route infeasible, so this figure directly limits how routes can be drawn. The model also records longitudinal and vertical load factors to assess passenger comfort.
Planning tool
I turned the airspace layers and the kinematic model into an open-source web tool, so people who don’t write code can design and check routes.
The frontend uses React. The 2D planner is a Leaflet map overlaid with approach corridors and Class B boundaries, switchable between north, south and west flow, with waypoints placed and dragged directly on the map. The backend is a FastAPI service running the kinematic model, written with NumPy and Pandas. It recalculates on every route change and marks any waypoint that makes the route infeasible. The 3D replay uses CesiumJS with free, onboard chase and tower cameras; the tower camera sits in the DTW tower cab and shows what a controller would see. Altitude and groundspeed profiles are plotted alongside in Plotly. The whole system is packaged with Docker and runs locally or on a server.
An FAA airspace specialist can use the free camera to check separation between a route and the approach corridors, and a human factors researcher can use the tower camera to find the approach directions that are hardest for a controller to see.
The rest of the paper
The paper also looks at demand and economics. Population growth regressions estimate the airport’s passenger base. Assuming 2% of airport trips shift from driving to air taxis by 2035, at $3 per mile and six seats per flight, ticket revenue reaches about $22 million a year by 2035. A round trip costs about $150 from Ann Arbor and $114 from Detroit.
An input-output (Leontief) model regionalized to Wayne County shows regional industry output growing by about $7.02 million by 2035 at $3 per mile, while sectors such as airport parking shrink as fewer travellers drive. At around $6.04 per mile every sector comes out non-negative, and regional output grows by about $44.13 million.
The approach corridor layers in the planning tool come from another part of the paper, where a Python program parses DTW’s approach procedures from the FAA NASR dataset and rebuilds 3D corridors in Google Earth for south, north and west flow.
The team also built an immersive DTW tower view in Unity with Cesium for Apple Vision Pro, intended for future studies of controller workload, how easily eVTOLs can be spotted, and whether AI decision aids help. Compatibility issues between Cesium and Vision Pro have so far kept it off the headset.
Planned next steps are a battery energy model so long holds stay within reserves, vertical airspace boundaries enforced inside the model, wind effects, automatic insertion of intermediate waypoints to respect the 884 m turn radius, and support for multiple aircraft at once.
The full paper is below.