Ornadyne's 1-Pound Flapping Drone Registered as a Bird at Edwards, in a Test With No Published Detection Rate
Hardware / news
Ornadyne's 1-Pound Flapping Drone Registered as a Bird at Edwards, in a Test With No Published Detection Rate
The four-person Y Combinator startup says other companies' sensors could not identify the O1 as a drone on September 17; Ornadyne has not said how many sensors or how many passes.

A flapping-wing drone built by Ornadyne, a Los Angeles startup with a team of four, registered as a bird on other companies' detection sensors during the Air Force's Game of Drones 26-3 exercise at Edwards Air Force Base on September 17, according to Task & Purpose. The result comes from the company's own account of the test; no sensor maker, detection rate or number of passes has been published.
The exercise was run by the 412th Test Wing's Experimental Test Force and the Defense Contract Management Agency's Blue List team, and the Mojave base hosts drone exercises each quarter, per Task & Purpose. It was the first time Ornadyne had flown against detection systems built by other private companies.
What the O1 is, by the numbers
The O1 is an ornithopter, a drone that flaps its wings instead of using propellers. Specifications below come from a company fact sheet quoted by Task & Purpose, and none has been independently measured.
| Specification | Ornadyne's figure |
|---|---|
| Weight | 1 lb |
| Range | Up to 10 km (about 6 miles) |
| Endurance | More than 1 hour |
| Payload | Small, such as a camera for target recognition |
| Wingspan and unit cost | Not stated |
The company says it can change the O1's colour, wingspan, proportions and silhouette to match a bird species native to the area it operates in. A 1 lb airframe with a camera-sized payload is a reconnaissance platform, not a munition carrier, and an hour of endurance at that weight puts it in the class of small hand-launched fixed-wing drones rather than the quadcopters that dominate current battlefield use.

The jet above is a T-7A trainer photographed over Edwards in 2023, shown for the venue and not the exercise or the drone.
What the founders said
Armen Arakelyan, who Y Combinator lists as founder and chief technology officer and who previously worked on Starship components at SpaceX, told Task & Purpose: "It looks like a bird, it moves similar to a bird, and humans and counter-UAS systems can misclassify it as a bird."
Geourg Kivijian, the founder and chief executive, previously at NASA's Jet Propulsion Laboratory and Astrolab, said "sensors deployed by the other companies here could not establish our drone using their detection equipment" and, of the result, "It registered as a bird, which is exactly what we wanted."
The Y Combinator listing for the Spring 2026 batch describes the product as "robot birds for surveillance" with "hour-class flight times" and "low acoustic signatures", and gives no figures.
What is missing
The test result is a single sentence of evidence. Task & Purpose does not report which sensors were used, whether they were radar, optical or acoustic, how far away the O1 flew, or how many times it was classified. A bird misclassification at long range by a radar tuned to ignore birds is a weaker result than one at 200 metres by an optical system trained on birds, and the account does not separate them.
There is also no operational data: no unit cost, no customer, no flight in wind or rain, and no figure for how noisy the flapping mechanism is, even though Kivijian says the drone is quieter than propeller designs. A one-hour endurance claim and a 10 km range are both company figures from a fact sheet.
The military question runs the other way too. The same behaviour that fools a counter-drone sensor is a problem for a base trying to tell a surveillance drone from a flock of crows. Our earlier coverage of Renault's Forefront ground robot and of Nominal's $240 million Series C shows how much of this market is still announcements and renderings.
What happens next
Ornadyne says it is pursuing demonstration partnerships with military stakeholders. The next evidence that would matter is a test with named sensors, repeated passes and a published classification count, run by someone other than the company.
Sources
More in Hardware
- 01Axon's $1.2 Billion Skydio Deal Puts Only $254.9 Million Into Skydio on Day OneThe headline sum is mostly a tender offer for existing holders' shares; the first tranche that closed sends fresh money to the drone maker in a single $254.9 million block.
- 02Agility Robotics Listed With $465 Million, Not the $620 Million Projected in June, After 61.5% RedeemedThe first U.S. pure-play humanoid company trades as AGLT on a $2.5 billion valuation, with $300 million of orders that one outlet says should not be read as booked revenue.
- 03Ford's 90% Battery Claim Rests on About 2,000 Mach-Es Past 100,000 Miles, and 30 Past 200,000Ford says 36,000 Mustang Mach-Es average 92% battery health, but neither the measurement method nor the data behind the lineup-wide figure has been published.
- 04Tesla's San Antonio Robotaxi Map Is 274 Square Miles on Paper, Probably 160 to 210 BookableThe advertised zone is four times Waymo's, but independent tests in six other cities found Tesla's apps accept pickups in 59% to 78% of the polygon.