Perch vs. hover: the energy math behind a 72-hour aerial sentry

The question#

Every multirotor security drone on the market quotes a flight time between roughly 20 and 55 minutes, whether it weighs two kilograms or forty. Customers reasonably ask why a bigger drone with a bigger battery doesn't simply fly for longer. The answer falls out of a page of 1920s rotor theory, and it is the reason we built Kestrel to land rather than hover. This note derives the result, puts numbers on it, and publishes the model (perch_vs_hover.py) so anyone can check it.

Hover power from momentum theory#

Treat the rotors as an actuator disk of total area $A$ pushing air downward. In hover, thrust equals weight, $T=mg$. Momentum theory gives the induced velocity through the disk and the ideal power to produce that thrust:

Real rotors lose power to blade drag, tip losses and non-uniform inflow. The figure of merit $\mathrm{FM}$ — ideal power divided by actual rotor power — captures this; small propellers typically land between 0.5 and 0.7. Motors and speed controllers add an electrical efficiency $\eta$. Battery power drawn in hover is then

where $\mathrm{DL}$ is disk loading in N/m². The second form is the useful one: power per newton of weight depends only on disk loading, air density and efficiency.

Worked example: a Kestrel-class airframe

With $m=41\text{kg}$, four 0.80 m rotors ($A=2.011{\text{m}}^2$, so $\mathrm{DL}\approx 200{\text{N/m}}^2$), sea-level $\rho =1.225$, $\mathrm{FM}=0.65$ and $\eta =0.85$:

Thirty-eight minutes of hover therefore needs about $4.2\text{kWh}$ of usable battery energy. That is the battery we carry.

Why bigger drones don't hover longer#

Write battery energy as a fraction $f_b$ of take-off mass times usable specific energy $e$ (J/kg). Endurance is energy over power:

Mass cancels. At a fixed disk loading and battery fraction, hover endurance is independent of size. Designers can stretch it by lowering disk loading (bigger rotors, which gets awkward fast because $t\propto {\mathrm{DL}}^{-1/2}$), raising the battery fraction (which eats payload) or improving cell chemistry. With $f_b=0.35$, $e=180\text{Wh/kg}$, $\mathrm{DL}=200{\text{N/m}}^2$ the formula gives $23.6\text{min}$ — the flat white line in the chart below, identical from 2 kg to 45 kg.

Perched endurance behaves completely differently. Once the aircraft is sitting on a ridge, the rotors are off. The load is the sensor suite, radios and housekeeping, which does not scale with airframe mass. Endurance becomes $t_{\text{perch}}=E/P_{\text{sensor}}$, and a heavier airframe simply carries more hours.

Results: perched endurance by sensor load#

For the 41 kg reference airframe and its 4.2 kWh battery:

The published 72-hour figure for Kestrel implies an average perched draw of at most $4166/72\approx 58\text{W}$ before solar input. A cooled thermal core, a solid-state radar in a low-duty search mode, the EO camera, mesh radio and compute fit inside that budget only with aggressive duty-cycling; an always-on high-power radar would not. That trade — search cadence against days on station — is a real engineering lever, and it is set per mission in Meridian rather than fixed at the factory.

Hot and high: density altitude#

Hover power scales as ${\rho }^{-1/2}$, so thin air costs endurance directly. Perched endurance is unaffected (sensors don't care about air density), but every relocation costs more:

A desert site at 2,000 m in summer loses about 14% of hover endurance. For a drone-in-a-box that is 14% less time on station per sortie; for a perching sentry it is 14% less transit range per battery, which matters only if relocation distances are already near the limit.

The cost of moving: duty cycle of a relocation#

Relocation is where a perching sentry goes blind, at least in our conservative model. A move costs a 9 s lift, a transit at 22 m/s and about 30 s to survey and set anchor feet. The fraction of time on watch with a relocation every $T_r$ seconds is

Even moving 3 km every hour keeps the unit watching 95% of the time. In practice the EO/IR gimbal keeps observing in transit, so these figures understate coverage; we report the conservative number.

Solar on the perch#

A perched airframe has a large, sun-facing top surface doing nothing. With a flexible photovoltaic skin of area $A_{pv}$, efficiency ${\eta }_{pv}$ and site-average daily insolation $H$ (kWh/m²/day), the average power it contributes is

With an assumed $A_{pv}=0.6{\text{m}}^2$, ${\eta }_{pv}=0.18$ and $H=4.5$ — a mid-latitude annual average, far lower in winter at high latitude — the skin adds about ${\bar{P}}_{pv}\approx 20\text{W}$. Against a 58 W sensor load that stretches 72 hours to roughly $4166/(58-20)\approx 110$ hours. The caveat is the denominator: as ${\bar{P}}_{pv}$ approaches $P_{\text{sensor}}$, endurance diverges on paper and becomes dominated by night-time depth of discharge, cloud runs and cell ageing in practice. We plan to 72 hours on battery alone and treat solar as margin, not as a capability.

The same arithmetic explains why solar is useless in the air. Hover draws 6.6 kW; a 20 W average contribution is 0.3%.

Limitations#

This is a first-order model. Momentum theory with a constant figure of merit ignores how FM varies with thrust and Reynolds number. Forward flight at moderate speed needs less power than hover, which flatters drone patrols slightly; the multicopter endurance literature (for example arXiv:2109.04741) treats this properly. We did not model battery temperature derating, ageing, reserve margins (a real operator keeps 20–30%), solar input on the perch, or wind loads while perched. The 15–120 W sensor loads are our assumptions spanning lightweight EO-only to radar-heavy payloads. None of these change the central result: hover endurance is capped by physics at tens of minutes, and perching removes the cap.

Frequently asked questions#

References#

1. J. Gordon Leishman, Principles of Helicopter Aerodynamics, 2nd ed., Cambridge University Press, 2006. Momentum theory, ideal hover power and figure of merit.
2. “Range, Endurance, and Optimal Speed Estimates for Multicopters,” arXiv:2109.04741. arxiv.org/abs/2109.04741

Reproducibility: model code is published at /research-code/ with a README. Results are model outputs under the assumptions stated above, not field trials.