C-DRONE GUIDE · 5 SEPTEMBER 2026

Detect and avoid (DAA): how a drone operator proves it controls the mid-air collision risk

An operator preparing a first beyond-visual-line-of-sight mission expects the paperwork to be the hard part: operations manual, declaration, authorisation file. In practice, what most often stalls a corridor application is neither administrative nor legal — it is proving that the remote pilot will spot an air ambulance, a glider or a crop-spraying aircraft entering the operational volume, and manoeuvre in time. That demonstration has a name in unmanned aviation: detect and avoid (DAA). It does not come as a single box you buy: it is assembled block by block, and every block has a documented blind spot. This guide walks through those blocks one at a time, states plainly what each one does not cover, and gives the quantified thresholds the SORA methodology expects for each air risk class of your operational volume.

Published on 5 September 2026, reviewed on 5 September 2026 — regulations in force as of September 2026.

"Avoid any risk of collision": an obligation that falls entirely on you

The founding text fits in one line. Implementing Regulation (EU) 2019/947, at point UAS.SPEC.060 3)(b) of its annex, provides that during flight the remote pilot "shall avoid any risk of collision with any manned aircraft and discontinue a flight when its continuation may present a risk to other aircraft, people, animals, environment or property". The obligation is unconditional and, above all, one-way: no symmetrical duty falls on the pilot of the manned aircraft, who does not have to see you, give way to you, or even know you are there. In the open category the regulation states the same requirement through human means: keeping the aircraft in visual line of sight and performing "a thorough visual scan of the airspace surrounding the unmanned aircraft".

That asymmetry is not a regulatory whim; it is empirically grounded. The historical "see and avoid" principle — ICAO Annex 2, section 3.2, echoed by rule SERA.3201 — assumed a pilot on board, with a canopy, peripheral vision and trained scanning habits. A study by Ryan Wallace and Samuel Vance published in 2019 in the International Journal of Aviation, Aeronautics, and Aerospace measured what that principle is worth in the other direction, by flying a multirotor into the approach path of general aviation aircraft: pilots detected the drone in only 30% of approaches — 13.6% when it was hovering, at a mean range of 647 feet, and 50% when it was moving, at a mean 1,593 feet (see the study on Google Scholar). The authors stress that at those ranges the margin left for a successful evasive manoeuvre is very thin.

The operational conclusion is blunt: you cannot rely on manned aviation to avoid you. The entire deconfliction burden falls back on the operator — and that is exactly what the risk assessment methodology wants you to demonstrate.

Where DAA fits into SORA: air risk, ARC, strategic mitigation, TMPR

The SORA methodology, in its edition 2.5 published by JARUS on 13 May 2024 and the reference in France since late September 2025, handles mid-air collision risk in three successive steps. Step 4 assigns an initial air risk class — the ARC, from a to d — by sorting your operational volume into one of twelve aggregated encounter categories defined by five criteria: altitude, controlled versus uncontrolled airspace, proximity of an airport or heliport, flight over urban versus rural areas, and atypical or segregated versus typical airspace. To build that reasoning from a chart, our guide to airspace classes A to G, CTR, TMA and LTA covers the groundwork. The ARC is explicitly a qualitative classification of the rate at which a drone would typically encounter a manned aircraft in that volume — not a measurement of your particular section.

Step 5 lets you bring that class back down through strategic mitigations, applied before take-off. SORA distinguishes two families: operational restrictions the operator controls directly — volume boundaries, time windows, capped height — and the structure and rules of the airspace, controlled by the authorities or a service provider. This is where, and only where, geo-awareness, checking the activation windows of the military very low level network or coordinating with an aerodrome operator belong. Remember it, because the confusion is common: geo-awareness detects no traffic at all; it stops you entering somewhere. It lowers the probability of an encounter; it does nothing about one once it happens.

Step 6 is where applications stall. It assigns a tactical mitigation performance requirement — the TMPR — to cover the residual risk left after strategic mitigations: ARC-b calls for a low TMPR, ARC-c a medium one, ARC-d a high one, ARC-a none. SORA Annex D translates those levels into quantitative risk-ratio objectives: the tactical mitigation system must bring the ratio to at most 0.66 for ARC-b, 0.33 for ARC-c and 0.1 for ARC-d. One crucial and often-missed nuance: visual line of sight remains an acceptable tactical mitigation for every ARC. As long as your eyes, or those of a chain of observers, cover the drone, no quantified requirement applies — see our guide to EVLOS and the STS-02 scenario with a chain of observers. But the methodology states its own limit: it assumes an observer cannot detect traffic beyond 2 NM (roughly 3.7 km), and requires EVLOS communication latency between remote pilot and observers to stay under 15 seconds. Cross that line and Annex D applies — and it decides whether a BVLOS flight beyond visual line of sight is feasible long before the operations manual does.

The building blocks, one by one: what each one sees, what it misses

The scientific literature splits the field in two, and that split explains most blind spots. The review by Xiang Yu and Youmin Zhang published in 2015 in Progress in Aerospace Sciences distinguishes cooperative approaches, where the aircraft to be detected broadcasts its own position, from non-cooperative ones, where the drone or a ground station must see it without its collaboration — radar, camera, infrared, acoustic sensing (see the study on Google Scholar). The whole difficulty of a real application is that traffic along a French corridor mixes both, in proportions that vary section by section.

Observers. Low cost, immediate to deploy, no technical dependency. SORA accepts them for every ARC, but assumes a detection range of about 2 NM, degraded by haze, intruder size and closing geometry. Along a linear asset the chain quickly becomes the heaviest logistical item — and the first thing wind or land access interrupts.
ADS-B in, airborne or ground-based. The most cited and most misunderstood block. It only shows aircraft transmitting ADS-B out. In mainland France, the order of 5 July 2024 requires VFR transponder carriage only in class A, B, C or D airspace, along certain published routes and for night flights away from the aerodrome vicinity: in low-level class G airspace, outside a published transponder mandatory zone, nothing obliges a light aircraft to be visible. Worse, EASA's study on electronic conspicuity interoperability recalls that a Mode S transponder alone carries no position information: an aircraft on Mode S without ADS-B extension will simply not appear on your display.
FLARM and the SRD 860 band. This is the de facto equipment of the gliding, paragliding and part of the microlight community — precisely the traffic a rural corridor meets. The same EASA study details its characteristics: unprotected band, transmit power of 10 to 25 mW for FLARM, an ideal range of 3 to 20 km but extremely sensitive to obstruction, which makes an external antenna near-mandatory. SORA Annex D puts it bluntly: a drone protected by a transponder-based system "will not mitigate all encounter risks in an area where sailplanes equipped with FLARM are known to operate".
Ground radar and detection cameras. Non-cooperative, they see what transmits nothing: that is their one real advantage, and it is decisive. But a study by Josh Chang, Teresa de Jesus Krings, Brendan Ooi, Iryna Borshchova and Jeremy Laliberté, published in 2025 in the Journal of Air Transportation, shows through quantitative modelling that short-range non-cooperative detection systems, used as the sole tactical mitigation, have only a minimal effect on collision risk; the authors conclude that adequate mitigation requires combining long-range non-cooperative sensors, cooperative sensors, strategic mitigations and traffic coordination (see the study on Google Scholar).
The traffic information service. Implementing Regulation (EU) 2021/664 provides for it in article 11, alongside network identification (article 8), geo-awareness (article 9) and flight authorisation (article 10). It changes the picture because it comes with point SERA.6005(c), introduced by Regulation (EU) 2021/666 and applicable since 26 January 2023: in designated, uncontrolled U-space airspace, manned aircraft "shall continuously make themselves electronically conspicuous to the U-space service providers". It is the one configuration where the asymmetry finally reverses — but it assumes your section sits inside a designated zone, which remains rare; see our guide to U-space and the drone's digital airspace.

The conclusion EASA draws from its own inventory deserves quoting as is: electronic conspicuity devices are "mostly not interoperable with each other, meaning that aircraft may or may not be electronically visible to each other". The agency adds that electronic conspicuity "is not to be seen as a collision avoidance system": it feeds situational awareness, it decides nothing. The ADS-L protocol, which EASA is developing precisely to make these worlds compatible, is still being specified.

Why fitting the drone with ADS-B out is not the answer

The idea comes up in almost every kick-off meeting: since the problem is that nobody sees the drone, let us fit it with a transmitter. It is a false good idea, for three cumulative reasons.

The first is frequency saturation. ADS-B out broadcasts on 1090 MHz, the same frequency used by Mode S transponders and on which commercial aviation's TCAS depends. EASA's electronic conspicuity study is unambiguous: "a disturbance or excessive stress of the 1090 MHz frequency leading to a degradation of TCAS performance must be avoided under all circumstances". The European strategy is to migrate already-installed Mode S transponders to ADS-B — over 75% of the existing fleet would be easily upgradable according to the manufacturer survey in that same document — while avoiding additional installations in other user groups. Multiplying transmitters across thousands of inspection and delivery drones runs exactly the other way.

The second is that transmitting only protects you if someone receives. A light aircraft without ADS-B in, which remains the common case, will see nothing of your signal: you would have loaded the frequency without gaining a metre of separation. The same document sums it up: without an in function and a display, there is no traffic picture.

The third is that conspicuity and identification must not be confused. The direct remote identification drones have carried since 2024 broadcasts over Wi-Fi or Bluetooth at a range under 2 km, a benefit the EASA study calls "questionable" as far as manned aviation is concerned: it serves the authorities and ground oversight, not in-flight avoidance. Network remote identification, on the other hand, feeds traffic management systems — that is the right channel, and the one U-space generalises.

What does hold true, and what SORA explicitly recommends, is to work on "all features that may increase the detectability of the UA in the airspace": strobe lighting, airframe contrast, predictable trajectory, published activity. But the block that actually serves your safety is reception, not transmission: an ADS-B in receiver combined with an SRD 860 receiver, onboard or deported to the ground along the section, aggregated onto the remote pilot's display.

Detecting is not enough: deciding, commanding, executing — and what it costs on a corridor

The most common mistake in an application is to bet everything on the sensor. SORA Annex D breaks DAA into five sub-functions — detect, decide, command, execute, and the feedback loop — and sets objectives on each. On detection the expectation is quantified: spotting roughly 50% of aircraft present in the detection volume at the low level (ARC-b), relying for instance on real-time traffic tracking services, low-cost ADS-B in or FLARM receivers, or monitoring aeronautical radio; roughly 90% at the medium level (ARC-c), which implies ground-based radar or DAA, an ADS-B in receiver, an ATC separation service or a UTM surveillance service. At the high level (ARC-d), SORA points to the RTCA SC-228 or EUROCAE WG-105 standards — that is, certified aviation equipment, out of reach for most operators.

The next three sub-functions are the ones nobody budgets for:

Deciding requires a written deconfliction scheme: which tools detect, which criteria trigger the manoeuvre, what phraseology applies if detection comes from a third party. At the medium level, SORA expects the remote pilot to decide within 5 seconds of the incoming-traffic indication, plus an assessment of the human-machine interface factors that might prevent it.
Commanding concerns the command and control link: total latency between the remote pilot's input and the aircraft executing it must not exceed 5 seconds at the low level and 3 seconds at the medium level. On a corridor with patchy network coverage, this is often what fails — not detection.
Executing holds a pleasant surprise: at the low level, SORA considers that descending below the nearest trees or buildings, or below 60 ft AGL, within less than a minute constitutes sufficient avoidance. For a low-level linear inspection that is a realistic manoeuvre requiring no particular flight performance. At the medium level, by contrast, the methodology suggests capabilities of the order of 50 kt airspeed, 500 ft/min climb or descent rate and a 3°/s turn rate.
The feedback loop closes the cycle: for an assumed 3 NM alerting threshold, a 5-second update rate and 10-second latency are deemed adequate at the low level; at the medium level, intruder position and vector data must refresh within 3 seconds.

On top of this comes a robustness requirement that, in practice, draws the boundary of what is feasible: allowable loss of function of the tactical mitigation system is less than one occurrence per 100 flight hours at ARC-a and ARC-b — SORA notes that commercially available products meet this without quantitative analysis — less than one per 1,000 hours at ARC-c, and less than one per 100,000 hours at ARC-d, with a mandatory quantitative analysis. In other words: the right strategy is almost never to buy more sensors, but to bring the ARC down through strategic mitigations — shifting time windows, capping height, splitting the linear asset, coordinating with local users. That is also the conclusion of the modelling by Chang and co-authors: only a combination of means, never an isolated sensor, reaches the expected level.

On orders of magnitude, let us stay cautious. The file itself — SORA analysis, assessment of detection-tool effectiveness, deconfliction scheme — sits in the €1,500 to €6,000 excl. VAT range already documented in our BVLOS and EVLOS guides, the upper bound corresponding to instrumented applications. A day of STS-02 flying with two observers runs between €900 and €1,600 excl. VAT, each additional observer between €200 and €350 excl. VAT per day. On the hardware side, no public price list exists for professional DAA chains; the one useful public benchmark comes from the EASA study, which structures general aviation electronic conspicuity into three tiers — a portable solution around €200, an uncertified device around €1,500, certified equipment around €5,000. Those figures concern manned cockpits, not an operator's DAA chain: they give an order of magnitude for one receiving station, to be multiplied by the number of points along a section and supplemented by data aggregation, any traffic information service and its maintenance. Over time the dominant cost is never the box: it is the demonstration and keeping it current. To scope a corridor mission — overhead power line, railway, pipeline, extended agricultural perimeter — request a quote stating the section length, the airspace classes crossed and the aerodromes or helipads nearby: those three items decide the ARC, and therefore the budget.

Frequently asked questions

Is an onboard ADS-B receiver enough to fly beyond visual line of sight?

No, and not merely for administrative reasons. An ADS-B in receiver only shows aircraft transmitting ADS-B out. In mainland France, the order of 5 July 2024 requires a VFR flight to carry a transponder only in class A, B, C or D airspace, along certain published routes, and for night flights leaving the aerodrome vicinity: in low-level class G airspace, where most inspection corridors run, a light aircraft has no carriage obligation outside a published transponder mandatory zone. On top of that, a Mode S transponder alone does not broadcast a position. The question is therefore not "do I have a receiver?" but "what share of the traffic actually present along my section does that receiver show me?" — and that share is what the authority wants to see estimated.

Should the drone itself carry ADS-B out or a transponder?

No — for a light drone it is actually counter-indicated. Transmitting ADS-B out on 1090 MHz adds load to a frequency which, as EASA's electronic conspicuity study points out, must not be over-stressed because it would degrade commercial aviation's TCAS performance — something that "must be avoided under all circumstances". The European roadmap aims to migrate existing transponders to ADS-B, not to add transmitters in new user groups. Making yourself visible remains useful, but through the channels designed for it: network remote identification and, in U-space airspace, the digital services.

Does a chain of observers replace a DAA system?

Within SORA, visual line of sight is an acceptable tactical mitigation whatever the air risk class: as long as the drone stays covered by human eyes, no quantified performance requirement applies. But the methodology states its own limit: it assumes an observer cannot detect traffic beyond 2 NM, roughly 3.7 km, and that figure degrades with haze, aircraft size and closing geometry. As soon as the flight genuinely goes beyond visual line of sight, SORA switches to the quantified requirements of its Annex D and human eyes are no longer enough.

Request a free quote