Two hours of modelled drone deliveries over a 12 × 12 km patch of Houston. Each arrow is a drone (a ring while it hovers). Red marks are planned conflicts between drones of different companies.
Where Amazon flies decides it. Amazon told Richardson council it flies out at about 69 m (225 ft); its FAA filings for Florida, Detroit and Kansas City put the return trip anywhere in 85–115 m, which includes Zipline’s 101 m cruise. Only if Amazon came home near the 400 ft (122 m) maximum in its filings, a height above every FAA-filed return band (at most 377 ft) that only an informal Amazon staff remark supports, would the benefit fall to about 29%. The 72% (95% interval 67–76% across 10 random layouts of Wing and Zipline sites; it does not cover model error) assumes 10 m of height between companies is enough; if 15 m were needed it would be 55%. The headline uses the filed Houston site counts, with Amazon’s real sites and Wing and Zipline sites placed at random; the animation below is a denser test patch: sites every 3 km, about 7 times as tightly packed as the filed layout, so it shows roughly 40 times more planned conflicts per square km than the headline study (120 in 2 hours over 144 km² against 45.8 an hour over 4,624 km²).
A model of a possible future build-out, not today’s traffic. With only the sites flying in Houston today, the companies’ plans almost never meet (HOUTODAY: 1–2 planned conflicts in 60 simulated hours); the question is what happens as they build out. Five of Amazon’s six Houston sites are real, geocoded locations (one, STX5, only to street level, up to about 1 km off); Wing’s and Zipline’s are placed at random because their full site lists are not public. The model flies planned paths only; it does not simulate collisions or avoidance. A “planned conflict” means two planned paths of different companies within 30 m sideways and less than 10 m vertically, with no coordination between companies. Some operators already share flight plans before take-off (the industry standard is ASTM F3548; Wing and Flytrex report doing so in DFW), so this is a reference point for how much height sharing removes, not a picture of today’s practice. The shared-out layers are 10–20 m apart, so the cut assumes 10 m between companies is enough (see Results for 15 m and 20 m).
Personal showcase. A personal, unaffiliated simulation project, built for learning and as a portfolio piece, using only publicly available FAA filings and public reports. Not affiliated with, endorsed by or reviewed by Wing, Zipline, Amazon or the FAA. Results come from a model, not real flight data, and are not safety, operational or legal advice.