Start here, because it reframes everything else: the FAA's own position is that DFR is a concept, not a special set of flight regulations. There is no DFR rule, no DFR certificate, no DFR application. What you are building is an operating model assembled from four separate things — a Part 91.113 Certificate of Waiver/Authorization (or, if your agency fails the public aircraft operator test, a Part 107 shielded-operations approval), launch infrastructure, a staffed remote operations center, and a written policy. The FAA only cares about the first.
The change that made this viable at scale was not a new rule either. It was one line in the standing conditions of the expedited Part 91.113 CoW/A the FAA stood up in April 2025: visual observers are "encouraged but not required." VO staffing had been the dominant recurring cost in a BVLOS program. Removing it turned DFR from a well-funded pilot project into something a mid-sized agency can afford.
DFR is a concept, not a rule: what you are actually assembling
The FAA's framing. If a vendor offers to get you "DFR approval," ask which instrument they mean. For most agencies there is exactly one: a Certificate of Waiver under 14 CFR 91.113.
The regulatory foundation, briefly
The instrument is a Part 91.113 CoW/A, processed by AFS-750 in Flight Standards, applied for by email to 9-AVS-AFS-750-91.113Waivers@faa.gov, valid 48 months. It replaced the Tactical BVLOS and First Responder BVLOS COAs and carries no NOTAM requirement and no monthly reporting.
It comes in two flavors. About 87 percent of applicants take the obstruction-shielded method: BVLOS to 200 ft AGL in Class G, or 100 ft above an obstruction within a 100 ft radius not to exceed 400 ft, at 1 SM visibility clear of clouds. The other 13 percent equip a detect-and-avoid system for 400 ft AGL at 3 SM visibility. DAA buys capability at the price of real integration work; most DFR programs launch shielded and stay there.
Eligibility is the gate that stops programs, not the paperwork. You must be both a public aircraft operator under 49 U.S.C. § 40102(a)(41) and a public safety organization under 49 U.S.C. § 44806(e). The FAA's own flow chart says flatly that many public safety organizations do not qualify; volunteer fire departments, SAR teams, special districts and university programs frequently fail. The fallback is Part 107 PSO Shielded Operations through AVSHub, a near equivalent limited to Class G. The package and the reasons submissions get returned are in the Part 91 COA and waiver guide.
The trap that grounds programs after approval. The 91.113 CoW/A bundles airspace authorization only up to the UASFM grid altitude or 200 ft AGL, whichever is lower. It is not unlimited airspace access. Anything above that, or controlled airspace with no active UASFM grid, needs a separate ATO COA through CAPS, filed after the waiver issues and at least 60 business days ahead.
The approval landscape as of 2026
The FAA published 985 Part 91.113 waivers issued as of its 31 July 2026 page update. Separately, an EFF FOIA release showed more than 1,000 agencies holding Part 91 BVLOS waivers as of February 2026, and the FAA issued more in the ten months after April 2025 than in the preceding seven years combined. Other counts circulate (976, roughly 1,400, 214 by June 2025) and do not reconcile, because they count different instruments over different windows. Do not quote a precise national total; those two attributed figures are the defensible ones. Processing runs 2 to 3 days. The bottleneck in 2026 is not the FAA.
The precedents worth knowing
Pearland, TX received a BVLOS waiver in March 2023 permitting operation without visual observers, the first of its kind. NYPD received a widely-cited approval announced 23 September 2024 covering much of the city, also without VOs and notably without a radar or electro-optical detect-and-avoid system, relying on shielded operations plus ADS-B In. Cite it if your ConOps proposes shielded operations in a dense urban core.
How to choose DFR launch sites and calculate coverage
Start with call-for-service data, not with buildings. Pull two to three years of CFS records, geocode them, filter to your intended call types, and build a heat map. Launch sites go where the qualifying calls are, weighted by time of day. Programs that site drones where the real estate was easy end up covering the parking lot behind headquarters.
Elevation is the second criterion. A rooftop site starts the aircraft above most obstructions and improves the shielded-operations geometry you are relying on. Each site needs shelter, reliable power, a stable network with enough uplink for live video, and environmental monitoring so the ROC knows wind at the pad rather than at an airport ten miles away.
Rooftop rights are the schedule risk nobody budgets for: on a building your agency does not own, expect a lease, roof-penetration approval, insurance certificates, and building-engineer access. Water towers, fire stations, public works yards, and libraries are the usual first wins because the property is already in the family.
The coverage radius arithmetic
Coverage is straight-line travel time plus launch latency. Do the arithmetic with your platform's real cruise speed rather than accepting a marketing radius. The waiver caps speed at 87 knots (100 mph).
| Cruise speed | 1 mile | 2 miles | 3 miles | 4 miles |
|---|---|---|---|---|
| 45 mph | 80 s | 160 s | 240 s | 320 s |
| 60 mph | 60 s | 120 s | 180 s | 240 s |
| 100 mph (max) | 36 s | 72 s | 108 s | 144 s |
These are transit figures in still air on a direct path. Add your measured launch latency (dock opening, power-up, GPS acquisition, climb), plus wind, routing, and the seconds before the launch decision. It runs longer than the spec sheet implies, and knowing by how much is what lets you site the next dock.
Docks and the charging tradeoff
A dock ("drone in a box") gives you an unattended, weather-protected, remotely launched aircraft. The decision inside it is charging time versus battery swap. A charging dock is simpler and cheaper, but the aircraft is unavailable for the full recharge cycle after every mission. A swap dock returns it to ready in minutes at the cost of moving parts and more failure modes. Call volume decides it: a high-tempo site finds its second call of the hour has no aircraft, and there is no software fix for that.
CAD and dispatch integration: put the launch decision inside the workflow
The technical requirement is an API interface to your CAD system plus Live911 call audio in the same console, so the teleoperator sees the incident record and hears the caller without leaving the screen they fly from.
The operational requirement determines whether the program works. The deploy decision has to happen inside the dispatch workflow, not beside it. If launching requires a dispatcher to notice the call, decide it qualifies, then reach a separate unit by radio, you have inserted a human relay into the exact seconds the program exists to save, and it will be skipped under load. Build it so the qualifying call surfaces automatically in the ROC console and the teleoperator launches or declines.
Radio integration matters for the same reason: the ROC belongs on the responding units' talkgroup. Some platforms carry exclusive integrations with specific radio, 911, and CAD stacks, which is both a legitimate sole-source ground and a constraint on your next procurement, covered in the agency drone procurement guide.
Staffing the remote operations center
Miami Beach PD's published model is one sworn officer plus one professional staff member per shift. That split reflects the core lesson of the discipline: separate piloting from radio duties. One person flies; the other works the radio, watches the CAD record, and coordinates with responding units. One person attempting both degrades at exactly the moment the call gets complicated. Whether the pilot must be sworn is an agency policy call, not an FAA one.
Staffing is the largest recurring cost line in a DFR program. Not the aircraft, not the docks, not the software subscription. Model it across every shift you intend to cover, including relief for leave and training, before committing to a coverage window. Going from 12-hour to 24-hour coverage roughly doubles the biggest number in the budget.
Call-type criteria and triage rules
Write the qualifying call types down before launch and approve them at the same level as any other use-of-resource policy. Good criteria weigh whether an aerial view arrives usefully faster than a ground unit, whether it changes the ground response, and whether the call type carries a scene-safety benefit. Alarm calls, area checks, vehicle collisions, brush and structure fires, missing persons, and in-progress property crimes are common inclusions.
Two restrictions belong in the policy explicitly, both drawn from the CJTTEC research: restrict use to active calls for service, and prohibit random patrol. An aircraft launching without a call attached is a patrol drone, a different program with a different legal and political profile, and blurring the two is the fastest way to lose the one you have.
When two qualifying calls arrive together and one aircraft is available, the teleoperator needs a written priority order, not a judgment call they will be second-guessed on later. Rank by threat to life, then by whether the aerial view changes the outcome, then by ground-unit ETA. Log the declined call, because declines are data about where the next launch site goes.
Drone first responder response times: what the data actually shows
Attribution matters here more than anywhere else, because these figures come from three different classes of source: agency self-reporting, a federally funded neutral evaluation, and vendor marketing.
| Figure | Source and status |
|---|---|
| 25,000+ missions; 17,170 drone-first arrivals; 96.98 sec average when drone-first; 4,138 arrests assisted; 4,629 calls cleared with no ground unit | Chula Vista PD's own published data, 8 May 2026 (launched 22 Oct 2018, five sites) |
| Drone first on scene on 72% of calls; 70-second average; drone response estimated at ~10% the cost of dispatching an officer and vehicle | Brookhaven, GA, agency-reported via CJTTEC |
| 62% of downtown-zone calls reached within 2 minutes | Asheville, NC, agency-reported via CJTTEC |
| Ground units returned to service in 25% of responses before arriving on scene | Cross-program figure, CJTTEC report |
| "86 seconds with a 4-mile coverage radius" | Flock Safety marketing claim, not audited data |
| Responder reaches scenes in about 70 seconds | BRINC marketing claim, not audited data |
The figure most worth understanding is not the 96.98 seconds. It is the 4,629 calls cleared with no ground officer response, about one in five missions. Response time is the headline; disposition without a patrol unit, and the cross-program 25 percent returned to service, are the fiscal argument.
The best neutral implementation source is *Parsons, N., Camello, M., Shute, R., and Planty, M., Drone as First Responder: Practical Insights into Law Enforcement Implementation, RTI International for CJTTEC, January 2026*, funded by the National Institute of Justice. Read it before writing your ConOps.
DFR policy, retention, and community trust
The policy elements the CJTTEC research identifies as standard practice are short enough to be non-negotiable: prohibit facial recognition, weaponization, and audio recording, restrict surveillance of constitutionally protected activity, maintain audit logs with restricted access, publish a dashboard of flight maps, incident types and aggregate metrics, and restrict use to active calls with random patrol prohibited.
Retention varies more than any other element, and there is no federal floor. Published practice runs from Montgomery County at about 180 days for non-evidentiary footage, to Chula Vista at 90 days unless flagged as evidence, to NYC at 30 days. Pick a number, justify it against your state records schedule and evidence rules, and publish it. An unstated retention period reads to the public as indefinite.
Eighteen states require a search warrant before deploying a drone for surveillance or evidence gathering, with exigency and rescue exceptions. Know whether yours is one before your first flight, not after your first suppression motion. California agencies carry a further obligation: drones are expressly "military equipment" under AB 481 with no size or cost carve-out, and the governing body must approve by ordinance before you solicit proposals, as the AB 481 guide explains.
Brookhaven PD sought feedback on its draft policy from the local district attorney and from the ACLU before launch. Worth copying: it costs a few weeks and converts the two organizations most likely to challenge the program into parties who saw the draft.
The critics, stated fairly
The ACLU (Jay Stanley, July 2023) called for a moratorium on drone-911-responder programs pending study, and where programs proceed, for deployment only in true emergencies, limits on video data handling, and disclosure of where and when drones fly, what sensors they carry, and what policies apply. The ACLU of Hawaiʻi formally opposed HPD's program and demanded independent oversight.
The EFF argues that drones reach backyards, roofs, and windows at distances leaving subjects unaware they are recorded, and that footage integrates with ALPR and other networks into something none of the individual systems was approved as. Their sharpest operational critique is the one program managers should sit with: deployments concentrate on low-risk calls involving unhoused individuals and mental health rather than emergencies.
That criticism is empirical, which makes it answerable, but only with measurement. Categorize every launch by call type and publish the distribution at a granularity where a skeptic can see whether welfare checks and disturbance calls dominate. If they do, that is a finding about your criteria and triage rules, better found in your own dashboard than in someone else's records request.
The DFR vendor landscape as of 2026
Unranked and without endorsement; verify current status directly, because this segment moves fast.
Flock Safety acquired Aerodome in October 2024 for over $300M and raised $275M led by a16z in March 2025, integrating DFR with its ALPR and camera network. Skydio offers the X10 and Dock for X10, is Blue UAS cleared, and partners with Axon. BRINC offers Lemur 2, Responder, and Guardian, launched March 2026 with Starlink; it raised $75M in April 2025 and $125M in July 2026, both led by Motorola Solutions, holds exclusive integrations with Motorola APX radios, VESTA 911 and CAD, and reports 900+ agency customers. Paladin Drones offered an $8,000 30-day DFR trial in 2026. DroneSense, Motorola Solutions, SkyFire Consulting, and Airdata also serve this market. DJI, historically dominant, is effectively closed to new US market entry after its December 2025 addition to the FCC Covered List; existing aircraft are not grounded, but software-update waivers run only through 1 January 2029.
A phased rollout: the first 90 days, the pilot, and what to measure
| Phase | Duration | Core work |
|---|---|---|
| Days 1–30 | Foundation | Confirm PAO and § 44806(e) eligibility; draft the ConOps; heat-map CFS data; pick candidate sites |
| Days 31–60 | Approval and policy | File the 91.113 package; draft policy (call types, retention, prohibitions); circulate to legal, the DA, and stakeholders |
| Days 61–90 | Build | Secure rooftop rights, power, network; stand up CAD and Live911 integration; train ROC staff; file an ATO COA if needed |
| Months 4–9 | Pilot | One or two sites, hours matched to peak CFS, a narrow call-type list; publish the dashboard from day one |
| Month 10+ | Expand | Add sites where declines clustered; widen call types on evidence; extend hours last, because staffing is the cost |
Because the application becomes part of the certificate, write the ConOps for the program you intend to run in eighteen months, not just the pilot.
Measure, at minimum: call creation to launch, launch to on-scene, share of qualifying calls where the drone arrived first, call-type distribution, calls resolved without a ground unit, units returned to service before arrival, declined launches with reason, and availability by site. Publish response times and call-type distribution together: the first is what the program is judged on, the second what it is criticized on.
Frequently asked questions
Do I need a special FAA approval to run a DFR program?
There is no DFR-specific approval. The FAA's position is that DFR is a concept, not a set of flight regulations. Agencies qualifying as public aircraft operators apply for a Part 91.113 Certificate of Waiver/Authorization through AFS-750, valid 48 months. Others use Part 107 PSO Shielded Operations, limited to Class G.
Do I still need visual observers for DFR?
No. Under the current Part 91.113 CoW/A, visual observers are encouraged but not required. That change, effective with the expedited waiver process the FAA stood up in April 2025, made DFR affordable at scale, because VO staffing had been the dominant recurring cost in a BVLOS program.
How long does the DFR waiver take to get approved?
Observed processing for the Part 91.113 CoW/A runs about two to three days, with one documented approval in 23 hours 47 minutes. The delay in standing up a program is almost never the FAA. It is rooftop rights, power and network, CAD integration, staffing, and policy approval by the governing body.
What are realistic drone first responder response times?
Chula Vista PD's own published data shows a 96.98-second average when the drone arrived before ground units, across 17,170 drone-first arrivals as of 8 May 2026. Brookhaven, GA reports a 70-second average, drone first on scene on 72 percent of calls. Vendor figures near 70 to 86 seconds are marketing claims, not audited data.
How much does a DFR program cost to run?
The largest recurring line is staffing the remote operations center, not hardware or software. Miami Beach PD runs one sworn officer plus one professional staff member per shift, which sets the floor for any coverage window. Other ongoing costs are launch-site power and network, dock maintenance, software subscription, and evidence storage.
Can a DFR drone patrol without a call for service?
Standard policy practice, reflected in the January 2026 CJTTEC research, restricts DFR use to active calls for service and prohibits random patrol. A drone launching without an attached call is a patrol program, with a different legal and political profile. Eighteen states also require a warrant for surveillance deployment, with exigency exceptions.
How long should we retain DFR video?
There is no federal retention floor, and published agency practice varies widely: Montgomery County retains non-evidentiary footage about 180 days, Chula Vista 90 days unless flagged as evidence, New York City 30 days. Choose a period you can justify against your state records schedule, then publish it. An unpublished period reads to the public as indefinite.
Where this leaves you
The approval is the easy part now: a qualifying agency can hold a 48-month Part 91.113 waiver in under a week. What separates programs that produce Chula Vista's numbers from programs that produce a press release is everything after the waiver — launch sites chosen from call data, a deploy decision inside the dispatch console, two people per shift with piloting and radio duties separated, written call-type criteria with triage rules, and a dashboard reporting call-type distribution alongside response time.
Four things in the next 30 days. Run the eligibility test against 49 U.S.C. § 40102(a)(41) and § 44806(e) first, because the answer determines which instrument you apply for. Heat-map your call-for-service data. Read the January 2026 CJTTEC report. And circulate the draft policy to legal counsel, your district attorney, and local civil liberties organizations before you file, the way Brookhaven did.
Flight records, currency tracking, incident documentation, and the flight-time totals the FAA can request at any time need somewhere more durable than a spreadsheet on a shared drive.