What “autonomous” means here#
An autonomous security drone launches, flies, observes and returns or lands without a pilot holding sticks. A person still supervises: approves the flight envelope, watches the feed when something happens and makes every consequential decision. In regulatory terms most of these operations are beyond visual line of sight (BVLOS), which is why regulation shapes architecture as much as engineering does (see drone security regulations).
There are four architectures in service. They differ in one property more than any other — the fraction of time the aircraft can actually see the site, which we call time on watch.
The four architectures#
| Architecture | How it watches | Time on watch, one point | Main constraint |
|---|---|---|---|
| Piloted / scheduled patrol | Flies a route on a timetable | Minutes per hour | Staff and battery |
| Drone-in-a-box (DiaB) | Launches from a charging dock on schedule or alarm | ≤ 30% per dock; ≈ flyable fraction with 4+ docks | Hover endurance, weather |
| Tethered drone | Hovers on a power cable above a ground unit | High while deployed | Fixed location, tether height, wind |
| Perching sentry | Flies to a vantage point, lands, watches, relocates | ≈ 99% in our weather model | Needs perches; transit windows |
The numbers in the third column come from our fleet sizing model. One DiaB dock with a 35-minute sortie and 55-minute recharge can cover a point for at most of the day; four docks with a well-designed scheduler reach the site's flyable fraction (91.1% in our synthetic year) and no more.
The endurance ceiling is physics#
Hover power follows from momentum theory:
Battery energy and hover power both scale with weight, so at constant disk loading and battery fraction, endurance is independent of size — about 23.6 minutes for typical parameters. That is why security drones from 2 kg to 40 kg all quote flight times in tens of minutes. Perching breaks the ceiling: the same 4.2 kWh battery that hovers a 41 kg airframe for 38 minutes runs a 58 W sensor suite for 72 hours. The derivation and results are in perch vs. hover.
Weather sets the upper bound#
Every flying architecture inherits the site's flyable fraction — the share of hours with wind, precipitation, visibility and icing inside the aircraft's limits. In our synthetic inland-site year with a 12 m/s wind limit and no flying in rain, 90.4% of hours were flyable. No DiaB fleet can exceed that. Architectures that watch without flying — tethered drones within their wind limit, and perching sentries — escape most of it. Get local wind and precipitation statistics before you compare vendors.
What a drone can see#
Elevation is why drones are used at all. But useful sight depends on resolution as well as line of sight: identifying a person needs about 250 pixels per metre on target under IEC 62676-4, and a 1080p camera with a 30° field of view delivers that only to about 14 m. Our DORI guide has the formula and tables. For line of sight, our viewshed study found that placement mattered more than height: two greedily placed masts saw as much as six random ones.
Regulation in one paragraph#
In the United States, routine BVLOS operations still require Part 107 waivers or other authorizations; the proposed Part 108 rule published in August 2025 would create a standing framework but, as of this writing, has not been finalized. Drones must broadcast Remote ID under 14 CFR Part 89. In the EU, most security operations fall into the “specific” category of Regulation 2019/947 and need an operational authorisation, typically supported by a SORA risk assessment. Details and sources are in our regulation overview.
What elevation adds to every sensor#
The reason to put sensors in the air is not the aircraft; it is the viewpoint. Three effects compound. Line of sight: in our viewshed study a sensor at 9 m saw 28.3% of the ground within 1.2 km against 13.9% at 2 m. Look-down angle: a camera looking down at 20–40° sees over parked vehicles and low walls that block a horizontal view, and thermal contrast against ground is usually better than against sky. Clutter: a radar above ground clutter sees low-flying drones that a ground radar loses behind buildings (see radar vs. RF).
Elevation also has a cost: distance. A camera 40 m up sees a person 50 m from the mast at about 64 m range, which matters for resolution at identification densities. Architectures that can move closer to a detection — patrol drones, docks responding to alarms and perching sentries — recover that resolution; towers cannot.
Failure modes seen in operation#
- Coverage claimed as availability. A dock “covering the site 24/7” means it can launch 24/7. Time on watch for one point is at most 30% per dock.
- Scheduler resynchronization. In our simulation, fixed-interval fleets drifted into flying together after weather holds and stayed near 28% coverage regardless of size.
- Weather blind spots that correlate with risk. Intruders choose rain and wind for the same reasons drones cannot fly in them. A system whose coverage collapses in bad weather is weakest when it is most needed.
- Battery logistics. Cells degrade with cycles and cold. A fleet sized on new-battery endurance loses coverage over its first year unless the operator plans replacements.
- Authorization scope. Waivers carry conditions — altitude, areas, crew — that quietly constrain what the system can do on the night.
Cost#
Commercial programs are mostly sold as a service. The comparison that matters is cost per hour of effective coverage, not per aircraft: a cheap drone on station 25% of the time is expensive coverage. Against guards, our Monte Carlo cost model puts a median 24/7 post at about $362,000 a year; see also the buyer's guide to security robot cost.
How to evaluate a system#
- Ask for time on watch for a specific point on your site, not availability to launch.
- Get your site's flyable fraction from local weather data and apply it to every flying option.
- Check resolution against DORI requirements at your actual distances.
- Ask how the scheduler behaves after a weather hold.
- Confirm the regulatory path for your jurisdiction and who holds the authorization.
- Price per hour of effective coverage, including monitoring.
Our RFP checklist turns these into vendor questions.
Frequently asked questions#
Can a security drone fly 24/7?
Not continuously in the air: hover endurance is tens of minutes. Continuous coverage needs several docked aircraft in rotation (limited by weather) or an aircraft that watches while landed, such as a perching sentry.
How long can security drones fly?
Typically 20–55 minutes. Momentum theory shows hover endurance is roughly independent of size at fixed disk loading and battery fraction.
Are autonomous security drones legal?
Yes, with authorization. In the US, BVLOS operations generally need Part 107 waivers pending the Part 108 rule; in the EU, most fall in the specific category under Regulation 2019/947.
References#
- FAA, “Normalizing Unmanned Aircraft Systems Beyond Visual Line of Sight Operations,” Notice of Proposed Rulemaking, Federal Register Vol. 90, No. 150, August 7, 2025. govinfo.gov
- 14 CFR Part 89, Remote Identification of Unmanned Aircraft; FAA Advisory Circular AC 89-1; ASTM F3586-22 (accepted means of compliance). faa.gov
- Commission Implementing Regulation (EU) 2019/947 on the rules and procedures for the operation of unmanned aircraft (open, specific and certified categories); JARUS SORA 2.5. easa.europa.eu
- IEC 62676-4:2014, Video surveillance systems for use in security applications — Part 4: Application guidelines. DORI pixel densities as summarised in Axis Communications, “Pixel density based on IEC 62676-4:2014.” whitepapers.axis.com