Writing about aerospace and electronic systems, particularly with defense applications. Areas of interest include radar, sonar, space, satellites, unmanned plaforms, hypersonic platforms, and artificial intelligence.
Suspicious aerial activity over Woomera signals the
Hypersonic Attack Cruise Missile may have reached one of its most
consequential milestones — while back home, cost growth and a compressed
test campaign continue to shadow America's premier scramjet-powered
strike weapon.
Aviation Week & Space Technology | Defense Staff | March 24, 2026
Bottom Line Up Front
The U.S. Air Force's Hypersonic Attack Cruise Missile — a
Raytheon/Northrop Grumman scramjet weapon intended to reach initial
operational capability in FY2027 — appears to have begun flight testing
at Australia's Woomera Range Complex this week under the bilateral
SCIFiRE program, consistent with FY2026 budget plans. The program
carries an estimated $2 billion development price tag, faces a reduced
test schedule of just five flights (down from seven), and is absorbing
projected cost overruns that prompted the Air Force to simultaneously
resurrect the rival AGM-183A ARRW boost-glide missile. Australia's RAAF
is integrating HACM onto its F/A-18F Super Hornets and will serve as a
critical test range partner throughout the campaign, operating under
AUKUS Pillar II framework.
Key Numbers:
Mach 5+Cruise Speed (to Mach 8 design goal)
~1,900 km Operational Range
$802 M FY2026 Budget Request
~$2BTotal Development Cost Estimate
5 Remaining Flight Tests Before IOC
FY2027 Planned IOC
Something unusual happened over the South Australian outback on the
weekend of March 22–23, 2026. A modified Gulfstream G550
intelligence-collection aircraft registered to the U.S. Missile Defense
Agency — tail number N551HA — transited from Hawaii through Guam to RAAF
Base Edinburgh near Adelaide, then began repeated high-altitude passes
along the boundary of the Woomera Protected Area, one of the world's
largest restricted overland test ranges at more than 120,000 square
kilometers. Flying in coordinated tandem was an Australian P-8A Poseidon
maritime patrol aircraft. Neither government confirmed or denied
whether a weapon had been launched. Both declined detailed comment.
To analysts who have tracked the Hypersonic Attack Cruise Missile
program, the pattern was familiar. The loitering flight profile of a
telemetry-collection platform, the concurrent use of a sensor-equipped
maritime patrol aircraft for overland surveillance coverage, the
deployment of MDA personnel to Edinburgh ahead of "sensitive" operations
at Woomera — all are consistent with the instrumentation and
range-clearance procedures associated with a hypersonic flight trial.
Sources confirmed to The Nightly, the Australian publication that first
reported the activity, that the flights were linked to work under the
SCIFiRE program — the Southern Cross Integrated Flight Research
Experiment — the bilateral U.S.-Australian framework that has
underpinned HACM's development and now provides the test infrastructure
the U.S. cannot replicate at home.
The Air Force will have time to conduct only
five flight tests before declaring the weapon operational — a reduction
from the original plan of seven.
A Weapon Born From Fifteen Years of Scramjet Research
HACM did not emerge from a clean sheet. It is the operational
successor to DARPA's Hypersonic Air-breathing Weapon Concept (HAWC),
which achieved successful powered hypersonic flights in March and July
2022 and January 2023 using a Raytheon/Northrop Grumman design. SCIFiRE
itself officially launched in November 2020 as an outgrowth of the 2007
Hypersonic International Flight Research Experimentation (HIFiRE)
program, which the same U.S.-Australian partnership used to explore
scramjet flight dynamics and reach Mach 8.
In December 2021 the Air Force approved HACM as a Middle-Tier
Acquisition rapid prototyping program, mandating completion within five
years. Raytheon received a $985 million cost-plus fixed-fee contract in
September 2022 to cover design, integration, qualification, and flight
testing of all-up rounds. A subsequent $407 million award in 2023 for
capability enhancements brought the total contract value to
approximately $1.4 billion. Northrop Grumman is responsible for the
scramjet propulsion system.
Under the SCIFiRE partnership, three manufacturers — Boeing,
Lockheed Martin, and Raytheon — submitted competing preliminary designs.
Raytheon was competitively selected in September 2022. "Under the
SCIFiRE partnership with Australia that was established in 2021, HACM
engaged three weapons manufacturers, executed three preliminary design
reviews, and competitively down-selected to Raytheon in September 2022,"
Dr. James Weber, the Pentagon's Principal Director for Hypersonics,
told Congress in written testimony. "This program recently conducted
wind tunnel testing of the all-up round and static fire ground tests for
its new rocket motor."
How HACM Works
HACM is a two-stage weapon. At launch from a tactical aircraft, a
solid rocket booster accelerates the missile to above Mach 4, sufficient
to sustain combustion in the Northrop Grumman scramjet engine that then
takes over for the cruise and terminal phases. Unlike boost-glide
designs such as the AGM-183A ARRW or the Army's Long-Range Hypersonic
Weapon (Dark Eagle), HACM sustains powered flight throughout its
trajectory, drawing atmospheric oxygen for combustion rather than
carrying oxidizer — an arrangement that reduces mass and enables the
approximately 1,900-kilometer range the Air Force specifies for the
system.
Speeds of Mach 5 to potentially Mach 8 — corresponding to roughly
6,200–9,800 km/h — characterize the cruise phase. The missile's compact
form factor, designed to be carried on F-15E Strike Eagles, F/A-18F
Super Hornets, EA-18G Growlers, F-35A Lightning IIs, and potentially the
P-8A Poseidon, represents a deliberate departure from larger
platform-constrained hypersonic weapons. A B-52 could potentially carry
20 or more HACMs; a B-1 could accommodate up to 36, according to Air
Force statements to Congress. This flexibility is central to the
program's rationale: dispersing hypersonic strike capacity across
diverse platforms rather than concentrating it on bombers.
The engineering challenges are formidable. Sustained scramjet
combustion — compressing supersonic airflow, mixing fuel, and
maintaining ignition within milliseconds — demands advanced thermal
protection as aerodynamic heating drives surface temperatures to extreme
levels. The challenge is roughly analogous to the thermal environment
of the SR-71 Blackbird, but sustained at operational altitudes with
precision guidance requirements that did not exist in the reconnaissance
era.
Schedule Slippage and Cost Growth
By June 2025, the Government Accountability Office's annual
assessment of major defense programs delivered a straightforward
verdict: HACM is "behind schedule." The preliminary design review
originally planned for March 2024 was deferred by six months to
September 2024 because, as program officials explained to GAO, more time
was needed to finalize the hardware design. A follow-on review to
certify the fully operational configuration for final flight tests was
rescheduled to sometime in 2025.
The cascading effect was significant. The Air Force will have time
to conduct only five flight tests before declaring the weapon
operational — a reduction from the original plan of seven. Air Force
officials told the GAO that five flights would still be sufficient to
establish confidence in the missile ahead of a rapid fielding decision. A
validation review covering the final configuration was expected to
precede the last test flights.
Cost growth has tracked alongside the schedule pressure. The
program's development cost as of January 2025 was estimated at close to
$2 billion — a two percent increase from the previous year's assessment
of $1.9 billion. More significantly, GAO reported that Raytheon is
"projecting that it will significantly exceed its cost baseline." Air
Force officials told the watchdog that eliminating two flight tests
could produce savings, but the FY2026 budget request — which funded HACM
at $802 million, up from $466.7 million appropriated in FY2025 —
suggests the Air Force chose to preserve the test campaign rather than
cut flights to control costs.
Air Force Secretary Troy Meink told lawmakers:
"We've got to be able to buy more than 10. We've got a big focus on
achieving scale and low cost."
An Air Force spokesperson declined to comment on the specific
status of HACM development citing "enhanced program security measures."
Raytheon did not respond to press inquiries. That reticence reflects a
deliberate classification posture: the service announced in early 2025
that it would withhold information on its hypersonics programs for
security reasons.
ARRW Resurrection — A Hedge Against HACM Delays
The Air Force's FY2026 budget request included a development not
widely anticipated: $387.1 million to resurrect the AGM-183A
Air-Launched Rapid Response Weapon, a Lockheed Martin boost-glide
missile previously deemed a lower priority than HACM. The service had
requested no ARRW funding in FY2025 after the program endured a rocky
test campaign that included a failed all-up-round test in 2023.
Air Force Chief of Staff Gen. David Allvin told House lawmakers on
June 5, 2025 that the service would pursue two distinct hypersonic
programs. "One is a larger form factor that is more strategic,
long-range, that we have already tested several times — it's called
ARRW. The other is HACM," he said. Air Force Secretary Troy Meink told
the same hearing: "We've got to be able to buy more than 10. We've got a
big focus on achieving scale and low cost for the weapons." The timing
of ARRW's restoration — concurrent with HACM's delayed first flight —
was widely interpreted as a hedge, ensuring the Air Force maintains a
hypersonic strike path even if HACM's development continues to encounter
friction.
The two weapons are structurally complementary. ARRW, boosted to
hypersonic speed by an ATACMS rocket before gliding to its target, is
larger and must be carried by bombers. HACM's air-breathing
configuration gives it longer range for its size and the ability to fly
"vastly different trajectories," as Air Force budget documents describe
it — a reference to the powered maneuverability that distinguishes
scramjet cruise from boost-glide.
Woomera: Why America Tests in Australia
The United States faces a structural constraint that the Woomera
partnership directly addresses: there is no domestic test range offering
the overland distance, remoteness, and airspace freedom required for
full-envelope hypersonic trials. GAO stated explicitly that "test range
availability and limitations in the U.S. have been an issue for
hypersonic programs" and identified the SCIFiRE/HACM integration as the
mechanism to alleviate it.
Australia's Woomera Protected Area covers over 120,000 square
kilometers of arid terrain in South Australia. Its restricted airspace
allows unimpeded flight paths exceeding 1,000 kilometers — a
near-requirement for a weapon with HACM's stated range. The range
infrastructure includes radar tracking stations, telemetry receivers,
and optical instrumentation capable of collecting aerodynamic and
thermal data from high-speed flight. Woomera has supported joint
U.S.-Australian hypersonic research since the HIFiRE program, providing
the instrumented overland corridor that no U.S. range can duplicate.
Australia's commitment extends beyond range access. The Australian
Defence Department confirmed that the RAAF is integrating HACM onto its
fleet of 24 F/A-18F Super Hornets, consistent with the 2024 Integrated
Investment Plan's intent to equip those aircraft with a hypersonic
weapon ahead of their planned retirement around 2040. "Through the
SCIFiRE agreement, the U.S. and Australia continue to collaborate on
HACM design and development, including efforts to integrate HACM on RAAF
F/A-18Fs and using Australian test infrastructure for flight tests," a
Defence spokesperson told Australian Defence Magazine. In Australian
service, HACM will be the first hypersonic weapon to be operated in the
Oceania region.
Australia's defence investment in the broader hypersonic domain is
substantial. The 2020 Defence Strategic Update included funding of $6
billion to $9 billion for high-speed long-range strike research out to
2040. In November 2024, the three AUKUS partners — the United States,
United Kingdom, and Australia — signed the Hypersonic Flight Test and
Experimentation (HyFliTE) Project Arrangement under AUKUS Pillar II,
creating a trilateral framework to share testing facilities and pool
technical expertise for both offensive hypersonic systems and
counter-hypersonic defenses.
AUKUS Pillar II: The Strategic Context
HACM's development is embedded within the larger AUKUS Pillar II
advanced capabilities framework, which encompasses eight technology
workstreams including hypersonics and counter-hypersonics, artificial
intelligence and autonomy, quantum technologies, and advanced cyber
capabilities. In April 2022 the White House announced the AUKUS partners
would work together to accelerate hypersonic development specifically
under this framework.
Congressional Research Service analysis notes that AUKUS Pillar II
appropriations for FY2025 included $69.8 million specifically for work
relating to AI, maritime hypersonic tracking and targeting, and
air-launched hypersonic cruise missiles — an almost certain reference to
HACM-related activity. Legislation passed as part of the 2026 National
Defense Authorization Act included the AUKUS Improvement Act to
streamline technology sharing, though analysts have noted that the State
Department's ITAR exemption failed to narrow the Excluded Technology
List as broadly as industry expected, creating continued friction in
technology transfer for high-end capabilities such as hypersonics and
unmanned systems.
The strategic rationale for urgency is not abstract. China
reportedly has five hypersonic missile series in testing or operational
use, and has conducted hypersonic tests at a pace U.S. officials have
described as vastly exceeding American test frequency. Russia has
claimed operational deployment of the Kinzhal air-launched ballistic
missile and is developing additional hypersonic systems. The United
States is developing three hypersonic weapons programs simultaneously:
the Air Force's HACM, the Navy's Conventional Prompt Strike (CPS) system
for Zumwalt-class destroyers and Virginia-class submarines, and the
Army's Long-Range Hypersonic Weapon (Dark Eagle). CPS achieved
successful end-to-end tests in June and December 2024 and again in April
2025 after early test failures, though its deployment to Zumwalt-class
destroyers has also slipped from FY2025 to 2027.
The Road to FY2027
The Air Force's current plan, as of this writing, calls for HACM to
achieve initial operational capability in FY2027 under the rapid
fielding phase of its Middle-Tier Acquisition program. The 13 prototype
rounds funded under the rapid prototyping effort serve as test assets,
spares, and a residual operational capability. A subsequent major
capability acquisition pathway program would begin production in FY2029,
informed by the data gathered during the prototype campaign.
To prepare for that transition, the FY2026 budget includes funding
specifically for Manufacturing Capacity Enhancements to ensure the
industrial base can handle a ramp to full-rate production. The Air Force
has revised its transition strategy to prioritize delivering more
missiles sooner, improving manufacturability of the design, and
expanding production capacity — changes driven by Secretary Meink's
explicit directive that any U.S. hypersonic arsenal must be scalable
well beyond token quantities.
The presence of an MDA telemetry aircraft over Woomera this week,
if it confirms an initial flight test, would mark a milestone long in
coming for a program that has faced more scrutiny than its classified
profile might suggest. Whether the test — if it occurred — was a
propulsion demonstration, a full all-up round flight, or an
instrumentation calibration sortie remains unknown. What is clear is
that the bilateral machinery underpinning America's most operationally
flexible hypersonic weapon is in motion, and the test range for doing so
sits not in Nevada or New Mexico, but in the red desert of South
Australia.
For a weapon that has traversed fifteen years of scramjet research,
survived the cancellation of its predecessor, and absorbed the
institutional turbulence of two billion dollars in development spending,
the next five flight tests may be among the most consequential in the
history of U.S. precision strike.
March 24, 2026 · Special Report: Autonomous Unmanned Systems
From improvised FPV bombs to AI-guided swarms,
mesh-networked sea drones, and autonomous ground robots, Ukraine's
three-year conflict has compressed decades of unmanned systems
development into a single, relentless test cycle — with global
implications that Western militaries are only beginning to absorb.
By Special Correspondent · Contributing Analysis:
Center for Strategic & International Studies (CSIS), Atlantic
Council, Lieber Institute (West Point), Hudson Institute, FPRI · March
2026
BLUF — Bottom Line Up Front
Ukraine's war with Russia has catalyzed the most rapid evolution in
unmanned combat systems since the invention of the cruise missile. Both
sides now deploy AI-guided aerial, maritime, and ground drones at
industrial scale, driving a shift from human-piloted UAS to unjammable,
autonomous platforms capable of collaborative swarming. Ukraine is
producing drones at a projected rate of five million or more per year,
has pioneered low-cost interceptor drones that are already deployed in
the Iran conflict, and has achieved historic first kills — including the
downing of two Russian Su-30 fighters by autonomous surface vessels.
Russia's Shahed campaign has scaled tenfold since early 2024 and is now
incorporating Nvidia AI chipsets and inter-drone mesh networking. Full
autonomous swarm capability is an inflection point approaching within
two to four years. Western militaries, by the assessment of leading
defense analysts, currently lag this conflict by roughly eighteen months
in operational readiness.
5M+Ukrainian drones projected for 2025 production
4,000+Russian Shaheds launched per month by Aug. 2025
1,500Ukrainian FPV interceptors produced daily as of Jan. 2026
I. From Consumer Drones to AI-Guided Weapons: The Compressed Arc of a Revolution
When Russian armor crossed the Ukrainian border in February 2022,
neither side possessed a mature doctrine for unmanned systems in
large-scale land warfare. What followed has been described by analysts
at the Center for Strategic and International Studies as less a gradual
evolution than an industrial arms race compressed into months — a cycle
in which concepts move from prototype to mass frontline deployment
faster than any Western procurement system is designed to accommodate.
"We count people, and we want our people to be as far from the front line as we can."
— Ukrainian military official, CSIS symposium, May 2025
Ukrainian troops initially repurposed commercial quadcopters for
battlefield reconnaissance. Within months, they had attached improvised
explosive devices to them and created a new class of low-cost attack
munition. Ukraine's battlefield experience reflects a shift toward
unmanned systems that augment or attempt to replace human operators in
the most dangerous missions, against an enemy willing to commit more and
more manpower to large-scale frontal assaults. The implications
cascaded rapidly: captured Russian soldiers reported not seeing a single
Ukrainian soldier on the front line — only drones.
Ukraine produced approximately 800,000 drones in 2023. By 2024, that
figure had grown to two million. In 2025, production targets reached
five million, with procurement expected to match. To appreciate the
scale, as CSIS Wadhwani Center Director Gregory Allen noted at a May
2025 symposium, U.S. missile procurement has historically been measured
in hundreds, or in exceptional years, low thousands of units. Ukraine is
operating in a different order of magnitude.
The strategic imperative driving this production surge is demographic
as much as technological. Ukraine is massively outnumbered. Autonomy is
described by leading Ukrainian developers as the single most impactful
defense technology of the century — because it transforms a manpower
challenge into a production challenge, which is far more manageable. As
Yaroslav Azhnyuk, founder of The Fourth Law — one of Ukraine's leading
AI drone companies — told IEEE Spectrum: once an operator can control
not one but twenty, fifty, or a hundred drones simultaneously, the
economics of the conflict change fundamentally.
II. The Architecture of Autonomous Navigation: How Jamming Forced an AI Revolution
The pivot to AI-guided autonomy was not planned — it was forced. By
2023, Russian electronic warfare (EW) had become profoundly effective
against human-piloted first-person-view (FPV) drones, severing radio
links and spoofing GPS receivers. According to the Royal United Services
Institute, Ukraine was losing approximately 10,000 drones per month,
mostly due to jamming. This attrition rate made human-in-the-loop
control, dependent on uninterrupted radio links, strategically untenable
at scale.
The response was to eliminate the link entirely. Ukraine's defense
industry has developed standalone AI-driven software that can be
integrated across various platforms to expand battlefield autonomy,
enabling environmental perception, target recognition, and navigation —
including last-mile approach to the target. This software comes in
standalone modules consisting of compact chips with embedded software
and sometimes cameras, which can be integrated into a range of platforms
from small FPV drones to long-range strike drones and turret-mounted
uncrewed ground vehicles.
Key Ukrainian AI Autonomy Systems — Status as of Q1 2026
The Fourth Law TFL-1: Terminal guidance module,
~$50 per unit, operational in 30+ Ukrainian military formations;
increases strike success rate up to 4× vs. operator-controlled drones.
First demonstrated in combat July 2025.
The Fourth Law TFL-2: Autonomous bombing
module. Operator designates target; AI locks on, calculates optimal
release point accounting for speed, wind, and altitude.
NORDA Dynamics Underdog: "Pixel lock" terminal
attack module. By summer 2025, fifth-generation software extended
autonomous lock-on range to 2,000 meters. Over 50,000 modules delivered
to frontline units.
Swift Beat (Eric Schmidt) / Bumblebee: AI
quadcopter with 70%+ direct-hit rate via autonomous terminal guidance;
jam-resistant visual-inertial odometry navigation; over 1,000 combat
flights by spring 2025.
Helsing HX-2 Karma: German AI-equipped UAV;
immune to EW through ability to search for, reidentify, and engage
targets without continuous data connection. First deliveries to Ukraine
December 2024.
Vermeer V.P.S.: AI visual positioning system
for GPS-denied deep-strike drones; deployed on Ukrainian long-range
strikes by summer 2025; U.S. Air Force contract for celestial navigation
variant. Raised $12M from Draper Associates.
Ukraine is pursuing an approach of training small AI models on small
datasets rather than developing large, all-encompassing models, enabling
fast and efficient onboard processing on the limited computing power of
small, inexpensive chips that can be quickly updated, retrained, and
upgraded to adapt to changing battlefield conditions. This "good enough,
fast" philosophy — a direct inversion of traditional Western defense
acquisition — has proven decisive.
What the New York Times documented after 18 months of frontline
reporting is the industrialization of autonomous drone warfare using the
same commercial technology stack that powers civilian operations:
visual positioning systems, AI target recognition, computer vision, and
even Raspberry Pi microcomputers.
— ◆ —
III. Russia's Shahed Evolution: From Cheap Loitering Munitions to AI-Enabled Mesh Networks
Russia's contribution to this revolution has centered on the Shahed
drone — an Iranian-origin design now mass-produced inside Russia as the
Geran-2. Originally a simple platform guided by inertial navigation and
GPS coordinates, the Shahed has undergone a systematic transformation
that has alarmed Ukrainian defenders and Western analysts alike.
"Now they are interconnected, exchanging information with each other.
They also have cameras allowing them to autonomously navigate to
objects. Soon they will be able to tell each other to avoid a jammed
region."
— Oleksii Solntsev, CEO, MaXon Systems, to IEEE Spectrum, 2025
Starting in September 2024, Shahed launches escalated sharply. Before
this period, the average weekly launch rate was around 130. Within six
months, the rate peaked at approximately 1,100 launches per week.
Despite Ukraine's continued success in intercepting or neutralizing
these drones with electronic warfare, the weekly number of successful
drone hits reached approximately 110 — nearly ten times higher than the
previous year's average.
The qualitative improvements are as alarming as the quantitative
ones. Between January 2024 and August 2025, the number of Shaheds and
Shahed-type drones launched per month increased more than tenfold, from
334 to more than 4,000. Ukrainian investigators found AI-enabling Nvidia
chipsets in Shahed wreckages, as well as thermal-vision modules capable
of locking onto targets at night.
Newer Shahed models use 4G data modems with Ukrainian SIM cards and
Chinese satellite navigation antennas, allowing them to navigate via
Ukrainian cell towers — a development that Kyiv's EU Ambassador
confirmed improves accuracy and complicates Ukrainian electronic warfare
defenses.
Russia is also fielding a more capable next-generation attack drone.
The V2U drone, used in strikes against the Sumy region, is outfitted
with Nvidia Jetson Orin processors and runs computer-vision software and
AI algorithms that allow it to navigate even where satellite navigation
is jammed. The sale of Nvidia chips to Russia is banned under U.S.
sanctions; press reports suggest the chips are reaching Russia via
intermediaries in India.
The China dimension is critically important. China supplies roughly
80 percent of the critical technologies used in Russian drones, and
engineers from both nations are collaborating closely on technology
development and battlefield adaptation. China leads the world in certain
AI applications, particularly computer vision and pattern recognition —
and Russian access to Chinese AI capabilities could narrow the
technological gap with Western systems faster than most Western analysts
currently anticipate.
According to President Zelenskyy's March 1, 2026 statement, Russia
launched over 14,670 guided aerial bombs, 738 missiles, and nearly
19,000 attack drones during the winter months of 2025–2026 alone. In the
final week of that period, Russia launched over 1,720 attack drones,
dropped nearly 1,300 guided aerial bombs, and fired over 100 missiles at
Ukraine.
IV. Ukraine's Counter-Drone Ecosystem: Drones Hunting Drones
Facing an adversary capable of overwhelming traditional air defenses
by sheer volume — and the impossibility of fielding enough
million-dollar interceptors against twenty-dollar targets — Ukraine made
a pivotal strategic choice in 2024: match cheap threats with cheap
counters.
After President Zelenskyy set production targets in July 2025,
Ukraine had, as of January 7, 2026, ramped up production to 1,500
FPV-based interceptor drones per day, designed specifically to counter
Shahed-type threats and other low-cost aerial targets. Interceptor
drones priced between $1,000 and $5,000 are being pitted against Shaheds
that, despite Russian efficiency gains, still cost $20,000 to $50,000
per unit. By the end of 2025, the average interceptor success rate had
reached 68 percent, according to President Zelenskyy.
A New York Times analysis found that Russia sent approximately 5,000
drones into Ukraine in February 2026, and Ukraine downed 87 percent of
them. This intercept rate has made Ukraine's expertise exportable and
urgently sought. When Iran began deploying Shaheds against U.S. and
Israeli targets in the Gulf in early 2026, Washington and its allies
found themselves acutely unprepared — and turned to Kyiv.
Ukraine dispatched drone interceptors and military personnel to
Jordan as Middle Eastern countries attempted to defend against Iranian
strikes, following a request from the United States. Zelenskyy confirmed
Ukraine's readiness to help, noting that "no other country in the world
has this kind of experience" with countering Shaheds — while
simultaneously requesting Patriot systems in exchange.
General Cherry Bullet/Sting/Octopus: Family of
FPV interceptors ranging from high-speed engagers to more autonomous
systems; Bullet maximum speed 280–300 km/h, ceiling 5,000m, endurance
7–10 minutes. UK partnering on "Octopus" variant: target output
2,000/month.
Skystriker (Kharkiv-based company): Fixed-wing interceptor drone capable of extended loiter to match Shahed flight profiles.
ODIN Win_Hit: Autonomous AI-based interceptor;
claimed capability against targets up to 800 km/h including cruise
missiles; AI handles detection, trajectory calculation, and engagement
without human operator.
Drone Wall (DWS-1 / Atreyd): Swarm coordination
system; single operator manages 100+ interceptors; AI automatically
allocates targets; combat testing began November–December 2025.
Project Eagle / Merops (Eric Schmidt): U.S.
startup largely autonomous interceptor system; over 1,000 Shaheds downed
as of November 2025; successful trials confirmed by Ukrainian military.
Ukraine embraces a "good enough" philosophy — rapidly fielding
inexpensive, effective systems to defend its population and territory as
quickly as possible. According to the Foreign Policy Research
Institute, despite multiple waves of attacks averaging hundreds of
drones per night, fewer than 10 percent of Shaheds manage to reach their
targets, and domestically produced interceptor drones now account for
nearly one-third of the Russian aerial threats successfully neutralized.
V. Silicon Valley Goes to War: The Private-Sector Acceleration
The scale of private-sector involvement in Ukraine's drone revolution
has no modern precedent. Hundreds of startups — many founded or staffed
by technologists who relocated from the United States and Western
Europe — are compressing commercial AI and robotics research directly
into combat hardware.
The most prominent figure is Eric Schmidt, former CEO of Google. In
July 2025, Ukrainian Defense Minister Rustem Umerov and Swift Beat CEO
Eric Schmidt signed a memorandum on long-term strategic partnership in
Denmark, in the presence of President Zelenskyy. The agreement covers
interceptor drones, reconnaissance quadcopters, and medium-class strike
drones — with production of hundreds of thousands of units projected for
2025 alone, with further increases in 2026.
Schmidt believes the outcome of future wars will be decided not by
the number of soldiers, tanks, or fighter jets, but by the autonomy of
systems and the power of algorithms. Ukraine is his testing ground for a
new technological revolution. Ukrainian military sources say Schmidt's
firm supplied three drone types responsible for downing approximately 90
percent of intercepted Russian Shaheds in those unit's engagements.
Schmidt's operation has cycled through multiple names — White Stork,
Project Eagle, Swift Beat. The Bumblebee quadcopter reportedly achieves
over a 70 percent direct-hit rate via autonomous terminal guidance,
autonomous target recognition highlighting foot soldiers, bunkers,
vehicles, and aerial drones before human pilots can spot them, and
jam-resistant navigation using visual inertial odometry.
Schmidt is not alone. Germany-based Helsing AI announced in December
2024 that the first of nearly 4,000 AI-equipped HX-2 Karma UAVs
earmarked for Ukraine were being delivered. The HX-2 is immune to
electronic warfare countermeasures through its ability to search for,
reidentify, and engage targets without a signal or continuous data
connection, while allowing a human operator to remain in or on the loop
for critical decisions.
The Fourth Law, founded by Yaroslav Azhnyuk, has dispatched more than
thousands of autonomy modules to troops in eastern Ukraine. The
company's TFL-1 terminal guidance and cruise modules are integrated with
dozens of manufacturers and continuously refined. Azhnyuk notes that
most frontline drones are expected to be fitted with similar autonomy
systems within six to nine months of early 2026.
— ◆ —
VI. Multi-Domain Warfare: Sea Drones Rewrite Naval History
The Ukrainian drone revolution has not been confined to the air.
Ukraine's unmanned surface vessel (USV) program — operating under the
Defense Intelligence Directorate's Group 13 — has achieved results that
maritime strategists are still processing.
Armed with AIM-9 Sidewinder missiles, Magura V7 drones shot down two
Russian Su-30 strike fighters in May 2025 — the first times in history
that fighter aircraft were downed by an uncrewed surface vessel. The
engagements, confirmed by Lt. Gen. Kyrylo Budanov, Ukraine's
intelligence chief, took place approximately 50 km west of Novorossiysk
in the Black Sea. The Magura V7 can conduct missions autonomously for 48
hours, or up to seven days when paired with a generator. Its payload
capacity is 650 kg, enabling simultaneous installation of a warhead,
machine-gun turret, and missile launchers.
The broader Black Sea campaign has been strategically decisive.
Ukraine's combination of Magura USVs with aerial FPV attacks forced the
Russian Black Sea Fleet to withdraw from its western positions and
retreat from bases near occupied Crimea, restoring effective Ukrainian
sea access despite the country having no surface warships of its own.
The implications were noted at NATO's REPMUS 2025 exercise in
September, where Ukraine brought upgraded Magura V7.2 drones to Troia,
Portugal, and served as the "red team" adversary, teaching NATO forces
what Russian tactics look like. The Portuguese Navy created its first
squadron-sized drone unit in 2023, directly inspired by Ukraine's
battlefield performance. By December 2025, Ukrainian President Zelenskyy
and Portuguese Prime Minister Luís Montenegro signed a joint
partnership for maritime drone production. Portugal joined the
Netherlands, Norway, the United Kingdom, Denmark, and Romania in
formalizing drone production collaboration with Ukraine.
VII. Ground Robots and the Emerging Unmanned Land Battle
As of early 2026, thousands of ground robots are operating across the
gray zone along the front line in Eastern Ukraine. Most are used to
deliver supplies or evacuate the wounded, but killer ground robots
fitted with turrets and remotely controlled machine guns have also been
tested. In mid-February 2026, Ukrainian authorities released footage of a
ground robot using its thermal camera to detect a Russian soldier at
night and neutralize the target with a heavy machine gun round.
Bryan Clark, senior fellow at the Hudson Institute's Center for
Defense Concepts and Technology, cautions that ground autonomy faces
distinct challenges relative to aerial platforms. Terrain complexity,
constrained sensor line-of-sight, and navigation difficulty in contested
environments mean that ground robot capabilities will advance more
slowly than aerial counterparts. The ultimate goal — one operator
controlling a mesh-connected swarm of autonomous ground systems —
remains aspirational for now, though developers assert their platforms
are already capable of basic autonomous operations such as returning to
base when radio contact is lost.
VIII. The Compute War: Infrastructure as a Strategic Variable
An underappreciated dimension of this conflict is the role of
computing infrastructure. Ukraine is producing drones at industrial
scale — well over three million annually across aerial, ground, and
maritime categories, toward a projected seven million in 2026. As
autonomy spreads throughout this ecosystem, bandwidth requirements will
outstrip available connectivity by orders of magnitude unless Ukraine
fundamentally restructures how and where computation occurs. Ukraine
operates approximately 58 data centers, compared with Russia's 251.
The Atlantic Council has modeled the risk scenario explicitly:
Russian EW assets severing tactical ground uplinks to Western cloud
infrastructure during a large autonomous swarm operation. In that
scenario, the swarm's ability to continue executing on preprogrammed
instructions — independent of real-time connectivity — becomes the
determining factor.
Russian access to Chinese AI expertise in autonomous systems, sensor
processing, and algorithmic targeting represents a strategic wildcard.
Chinese engineers from both nations are collaborating closely on
technology development, and this partnership could narrow the
technological gap with Western systems faster than current Western
analysis anticipates.
IX. Limitations, Risks, and the Contested Ethics of Autonomous Lethal Systems
Despite remarkable progress, leading analysts uniformly caution
against projecting current capabilities forward without accounting for
persistent technical, operational, and legal constraints.
On capability limits: while existing AI systems perform well
recognizing and following large objects like Shaheds or tanks, AI cannot
reliably distinguish a Russian soldier from a Ukrainian soldier, or a
combatant from a civilian. Tracking fast-moving infantry on motorcycles
and buggies remains "really challenging" for AI-guided systems.
Sensor quality is a binding constraint. Clark at the Hudson Institute
notes that AI navigation algorithms may be "pretty good," but they rely
on sensors that are not good enough. "You need multiphenomenology
sensors that can look at infrared and visual and, in some cases,
different parts of the infrared spectrum to determine whether something
is a decoy or a real target." Marc Lange, a German defense analyst, adds
that 2D image-based systems are too easily fooled: Russia demonstrated
this by drawing birds on the backs of their drones to confuse visual
recognition systems.
Cost remains a gating factor for full autonomy. The more autonomous
the system, the more expensive are the processors and sensors it
requires. For cheap attack drones that fly once, high-resolution cameras
and powerful AI chips are economically prohibitive. Until a balance is
achieved between technological sophistication and minimum cost, mass
autonomous deployment will be constrained.
Kate Bondar, formerly a policy advisor to the Ukrainian government
and currently a research fellow at CSIS, offers a measured
two-to-three-year timeline for "pretty good full autonomy, at least in
good weather conditions" for aerial systems — while emphasizing that
humans will remain in the decision loop for years, and full machine
autonomy without human oversight will not be operationally reliable for
at least a decade.
On the legal and ethical front: the "Stop Killer Robots" campaign has
urged states to push for new international law on autonomous weapons by
2026. But even if some states agree to halt development, China and
Russia will not stop their own efforts given the ongoing technological
arms race — and the Russo-Ukrainian War demonstrates that these systems
are already in use.
X. Strategic Implications: What the West Has — and Has Not — Learned
The consensus among analysts is stark: Western militaries remain
dangerously behind the operational reality emerging from Ukraine.
Yaroslav Azhnyuk of The Fourth Law puts it bluntly: while Russia and
Ukraine have made major strides over the past year, "Europe and the
United States have progressed, in the best-case scenario, from the
winter-of-2022 technology to the summer-of-2022 technology. The gap is
getting wider."
The United States and its Gulf allies discovered this gap
catastrophically when Iranian Shaheds — the same platform Ukraine has
been managing for three years — struck U.S. Navy infrastructure in
Bahrain and overwhelmed sophisticated Western air defenses. The U.S. has
now turned to Ukrainian expertise and interceptor technology as a
near-term remediation.
What happens on the battlefields of Ukraine can potentially define
how belligerents use military autonomy in other armed conflicts
globally. Nefarious actors have observed closely: FPV drones are already
being used by Islamic terrorist groups in Africa and by Mexican drug
cartels against local authorities. The proliferation trajectory of
autonomous lethal systems mirrors that of earlier disruptive weapons —
difficult to control once the knowledge and industrial base exists.
The implications for NATO's eastern flank are particularly acute.
Germany's Bundeswehr, which spent decades optimizing for industrial-era
warfare, is now engaged in an emergency re-orientation. European defense
agencies are studying Ukraine's rapid iteration model — from concept to
combat in weeks rather than years — as an aspirational standard. The
challenge is structural: European defense procurement cycles, liability
frameworks, and acquisition regulations are poorly suited to the OODA
loop required to compete in this environment.
Danish Prime Minister Mette Frederiksen summarized the emergent
geopolitical reality plainly at a December 2025 summit: "The only expert
right now in the world when it comes to anti-drone capacities is
Ukraine." The question for Western defense establishments is how much
time they have before they need that expertise operationally — and
whether they have been humble enough to absorb it.
— ◆ —
XI. Looking Ahead: The Inflection Points to Watch
Expert estimates from the Atlantic Council, Institute for the Study
of War, and UNITED24 suggest that by the end of 2026, AI-enabled
interceptors and swarm systems could down 40 to 50 percent of Shaheds
during mass attacks, with single-operator management of hundreds of
simultaneous drones.
Several technology thresholds will define the next phase of the
conflict — and by extension, the next phase of autonomous warfare
globally:
Passive radar maturity: Oleksandr Barabash of
Ukrainian startup Falcons identifies passive radar — which exploits
existing environmental signals from TV towers and radio transmitters
rather than emitting its own — as the critical counter to unjammable
autonomous drones. Unlike active radar, passive systems cannot be
targeted by anti-radiation missiles. Falcons received U.S. Green Flag
Ventures funding in September 2025 and is pursuing NATO certification.
Swarm coordination at scale: The transition from
individual autonomous drones to genuinely coordinated swarms — sharing
targeting data, avoiding intercepted zones, and adapting in real time —
is technically achievable with existing software but demands computing
infrastructure and mesh networking not yet mature at frontline scale.
Ukrainian developers are already testing AI-based mission planning using
simulations of thousands of combat scenarios, with the objective of
enabling fewer operators to manage larger numbers of coordinated
systems.
Quantum navigation: Ukraine is reportedly testing
quantum gyroscopes and accelerometers in 2026, in partnership with firms
like Vector Atomic, that would allow drones to navigate in total
electronic warfare environments — potentially making jamming-based
defenses obsolete.
Maritime swarms: The combination of proven Ukrainian
USV technology with American-developed swarm autonomy software — as
explored in the December 2025 HavocAI demonstration for Ukrainian
officials in Portugal — may produce a new class of distributed,
attritable naval strike capability with no precedent in existing naval
doctrine.
The war in Ukraine has become, as Eric Schmidt has stated repeatedly,
not just a conflict but a technology accelerator — compressing what
would otherwise take decades into an operational cycle measured in
weeks. The world's defense establishments are watching. The question is
which ones are learning fast enough.
XII. WW1 &WW2 Deja Vu All Over Again
**World War I** is the closer analog to where we are now. The first years of that war saw existing military doctrine collide catastrophically with industrial-era technology that outpaced tactical thinking. Machine guns, barbed wire, and artillery created a killing ground that neither side knew how to navigate. Improvised solutions proliferated at the front — often invented by soldiers, not general staffs. The tank was literally a hack, born of desperation to cross no-man's land, first deployed in 1916 before anyone had a mature doctrine for it.
Ukraine's FPV drone evolution mirrors this almost exactly. Troops strapping grenades to DJI Mavics because they had nothing else, then iterating from there — at a pace that left senior commanders playing catch-up.
**World War II** then represents what happens when the improvisations of the previous conflict get industrialized, systematized, and combined into new doctrine — blitzkrieg being the canonical example. The pieces (tanks, radios, close air support) all existed in WWI, but it took the interwar period to synthesize them into combined arms maneuver warfare.
The drone swarm with AI coordination is arguably that synthesis moment arriving now — the point where the improvised pieces (autonomous navigation, mesh networking, cheap sensors, mass production) are being integrated into coherent operational doctrine.
A few other parallels worth considering:
**The cost-exchange inversion** echoes the submarine warfare of both wars — a relatively cheap platform threatening assets that cost orders of magnitude more. Germany nearly starved Britain with a weapon the Royal Navy initially dismissed. The Shahed-versus-Patriot arithmetic has that same asymmetric logic.
**Electronic warfare as the new gas warfare** — a domain weapon that emerged mid-conflict, required constant adaptation, and made entire categories of existing equipment temporarily obsolete. GPS jamming has the same character: it didn't eliminate drones, but it forced a fundamental redesign of how they work.
**The convergence of civilian and military technology** has a WWI parallel too — the rapid militarization of aircraft, which were barely a decade old when the war started. The Wright Brothers flew in 1903; 15 years later, by 1918 there were strategic bombing campaigns. The commercial drone industry is on a similar timeline relative to the Ukraine war.
The most sobering parallel may be this: in both world wars, the powers that failed to absorb the lessons of early-war improvisation — and continued to fight with the doctrine and equipment of the previous era — paid for it in catastrophic casualties. The question Western defense establishments face right now is whether they are in the 1915 mindset or the 1940 mindset. The analysts surveyed in that article suggest the answer, currently, is closer to 1915.
6.
Bendett, S. & Kirichenko, D. "Ukraine Symposium – The Continuing Autonomous Arms Race." Lieber Institute for Law and Warfare, U.S. Military Academy at West Point, February 19, 2025.
https://lieber.westpoint.edu/continuing-autonomous-arms-race/
SAFETY | ACCIDENT INVESTIGATION | AIR TRAFFIC CONTROL
Breaking / Developing Story — Updated Throughout the Day
Fatal LaGuardia Runway Collision Lays Bare Systemic Gaps in Surface Safety Architecture
Two Jazz Aviation pilots are dead after an Air
Canada Express CRJ-900 struck a Port Authority ARFF vehicle on Runway 4
in the worst fatal accident at the airport in three decades — raising
urgent questions about ATC workload, ASDE-X alert performance, and
whether a decade of runway safety investment has been enough.
Special Report — Aviation Week & Space Technology Staff
Published: March 23, 2026, 15:30 ET
Updated: Ongoing — NTSB Go-Team On Scene
At approximately 23:40 ET on March 22, 2026, Air Canada Express
Flight 8646 — a Jazz Aviation Bombardier CRJ-900LR arriving from
Montréal — struck a Port Authority Aircraft Rescue and Firefighting
(ARFF) vehicle on Runway 4 at New York LaGuardia Airport (LGA), killing
both pilots and injuring 41 others. The ARFF truck had been cleared by
ATC to cross the active landing runway in response to a simultaneous
emergency involving a United Airlines aborted takeoff; the clearance was
rescinded seconds too late. LaGuardia — equipped with both ASDE-X
surface surveillance radar and FAA Runway Status Lights — was operating
under night VMC with mist and rain. The NTSB has deployed a Go-Team and
is leading the investigation with support from Canada's Transportation
Safety Board. ATC staffing levels at the time of the accident, Safety
Logic alert performance, and ARFF vehicle coordination protocols are all
under investigative scrutiny. The accident is the first fatal
commercial runway collision in the United States since the 2025 Reagan
National mid-air.
The Collision Sequence
Flight AC8646, a Bombardier CRJ-900LR registered C-GNJZ and
operated by Jazz Aviation on behalf of Air Canada Express, completed a
routine one-hour service from Montréal-Pierre Elliott Trudeau
International Airport (YUL) before touching down on Runway 4 at
LaGuardia at approximately 23:37 local time. Carrying 72 passengers and
four crew members, the aircraft was decelerating along the runway when
it encountered a Port Authority ARFF vehicle crossing at Taxiway Delta —
directly in the landing path.
Air traffic control audio recorded by LiveATC.net and ATC.com
captured the sequence with stark clarity. In the recording, the ARFF
vehicle radio operator transmits: "Truck 1 and company LaGuardia Tower requesting to cross 4 at Delta."
The controller issues the crossing clearance. Seconds later —
apparently realizing the Air Canada aircraft was still rolling on the
runway — the same controller is heard transmitting: "Stop, stop, stop, Truck 1, stop, stop, stop." The collision occurs before the truck can comply.
FAA Administrator Bryan Bedford confirmed that the ARFF vehicle
had been dispatched to respond to United Airlines Flight 2384, a Boeing
aircraft that had aborted its takeoff on the opposite side of the
airport after an anti-ice warning light illuminated and crew reported a
cabin odor that sickened flight attendants. The United crew declared an
emergency when no gate was immediately available. The ARFF vehicle was
authorized to cross Runway 4 at Taxiway Delta to reach the United
aircraft — placing it directly in the path of the decelerating CRJ-900.
FlightRadar24 preliminary data indicated the Air Canada
aircraft was traveling between 93 and 105 mph at the point of impact,
though a second Flightradar24 data point cited by Reuters placed ground
speed at approximately 24 mph at last contact. NBC News law enforcement
sources cited approximately 30 mph at collision. The discrepancy will be
resolved by the Flight Data Recorder, which the NTSB is analyzing.
"The two pilots who were killed were young men at the start of their career. This is an absolute tragedy."
The nose section of the CRJ-900 bore the full force of impact.
The height differential between the relatively low-slung regional jet
and the mass of the heavy ARFF truck concentrated crash energy at the
cockpit. Both the captain and first officer — neither yet publicly
identified — were pronounced dead at the scene. Forty-one people were
transported to Queens hospitals: 39 from the aircraft and two Port
Authority ARFF officers, both hospitalized in stable condition with
broken bones. CBS News law enforcement sources reported one passenger
suffered a traumatic brain bleed; a flight attendant, strapped in her
seat, fell through an opening in the severed forward fuselage. Nine
passengers remained hospitalized as of Monday afternoon, some with
serious injuries. An unaccompanied minor aboard was reunited with
family, Port Authority Executive Director Kathryn Garcia confirmed.
ATC Coordination and the Simultaneous Emergency Problem
The accident exhibits what aviation safety professionals
recognize as a classic concurrent-emergency failure mode: a controller
simultaneously managing two separate crises on opposite sides of the
airport field — a declared emergency on one runway and an active landing
on another — authorizing a crossing movement without adequate
deconfliction of the two operations.
Aviation safety analysts who reviewed available ATC audio noted
that the controller appears to have been managing both local tower and
ground control functions. Reports citing aviation experts suggest the
possibility that a single controller was working combined positions,
though Transportation Secretary Sean Duffy pushed back on this
characterization at a Monday afternoon press conference, stating that
reports of a sole controller were "not accurate." Duffy declined,
however, to specify how many controllers were on duty or whether
combined positions were in use. He confirmed that LGA has 33 certified
controllers against a target of 37 and seven controllers in training — a
shortfall of four certified controllers at one of the nation's busiest
airports.
The FAA confirmed to TIME magazine that ATC staffing levels at
LaGuardia at the time of the collision "will be part of the
investigation." NTSB Chair Jennifer Homendy and Member John DeLeeuw are
serving as on-scene spokespersons for the investigation. The
Transportation Safety Board of Canada has also deployed a team, as the
aircraft was registered in Canada and operated by a Canadian carrier.
The Air Line Pilots Association International (ALPA) dispatched
representatives to support the NTSB.
Investigative Focus Areas — NTSB Go-Team
Per available reporting and standard NTSB investigative
protocols for runway incursion accidents, the Go-Team's inquiry is
expected to examine: ATC staffing levels and position consolidation
practices at time of collision; ASDE-X Safety Logic alert status and any
generated alarms; Runway Status Light status at Taxiway Delta; ARFF
vehicle ADS-B/transponder equipment and multilateration tracking
coverage; ATC display configuration and controller situational
awareness; FAA Order JO 7110.65 runway crossing clearance procedures;
and weather conditions, including reported mist, rain, and standing
water on Runway 4.
Surface Surveillance Infrastructure: What Was There, What Should Have Triggered
LaGuardia is among the 35 U.S. airports equipped with the FAA's
Airport Surface Detection Equipment, Model X (ASDE-X) — a multi-sensor
fusion system that integrates surface movement radar, multilateration,
and ADS-B to produce a continuously updated positional picture of all
aircraft and vehicles on the airport movement area. The FAA's ASDE-X
program, the successor to the ASDE-3/AMASS (Airport Movement Area Safety
System) architecture deployed at major airports through the 1990s and
2000s, includes a conflict detection and alerting module known as Safety
Logic, which is designed to detect converging tracks and issue both
visual and aural alerts to tower controllers.
LaGuardia is additionally equipped with FAA Runway Status
Lights (RWSL) — an independent, automated pavement-embedded safety
system fed by the ASDE-X data stream that illuminates red Runway
Entrance Lights at taxiway-runway intersections when an aircraft is
detected as active on the runway. The RWSL system is specifically
designed to provide an automated, controller-independent warning
directly to ground vehicle operators at the stop bar — the precise
scenario that unfolded at Taxiway Delta on Sunday night.
ASDE-X is explicitly designed to track non-transponder-equipped
vehicles through its surface movement radar component — the radar
element provides skin-track returns from large metallic ground vehicles
regardless of whether those vehicles carry cooperative ADS-B or
transponder equipment. A Port Authority ARFF truck is a large, highly
radar-reflective target; at X-band or Ka-band surface surveillance
frequencies, it would generate a strong return. The fusion of radar
skin-track data with any available multilateration or ADS-B information
from the truck would then feed into Safety Logic's conflict detection
algorithms.
Those questions were answered at the NTSB's first full on-scene
press briefing Monday afternoon. NTSB Chair Homendy confirmed that the
FAA Technical Center's analysis of the ASDE-X replay is unambiguous: "ASDE-X
did not generate an alert due to the close proximity of vehicles
merging and unmerging near the runway, resulting in the inability to
create a track of high confidence." Homendy confirmed she reviewed
the replay herself and observed two radar blobs — skin-track returns
only, with no identification data — on Taxiway Delta. Neither target was
seen crossing in front of the aircraft on the replay. The Runway Status
Lights were reported as functioning per the replay, though the NTSB
stated that finding requires verification by FAA Technical Operations.
The root cause of the ASDE-X failure is direct and damning:
Truck 1 carried no transponder. Homendy confirmed there was no
indication that any of the ARFF vehicles involved carried transponders.
Without a cooperative track, ASDE-X's Safety Logic was operating on
radar skin-returns only. As multiple ARFF vehicles staged near Taxiway
Delta in the seconds before the crossing clearance — merging and
unmerging as the convoy prepared to move — the track-management
algorithms could not maintain a high-confidence track on any individual
vehicle. The Safety Logic conflict detection algorithm requires stable,
identified tracks to compute conflict geometry; with only ambiguous,
merging radar blobs, no alert fired. This is precisely the
nuisance-suppression failure mode that engineers on the original
ASDE-3/AMASS program identified decades ago: the system cannot reliably
discriminate between a threat and clutter when ground vehicles cluster
near runway thresholds without cooperative identification data.
The ATC Staffing Context
The accident occurred against a backdrop of well-documented,
chronic ATC understaffing that has been the subject of congressional
inquiry, union advocacy, and a formal National Academies of Sciences
study released in June 2025. That report found that ATC facility
shortages were attributable to past hiring constraints and a
misallocated workforce, compounded by inefficiencies in shift
scheduling, and that failure rates for achieving full certification at
individual facilities were increasing — particularly at large facilities
handling the most complex commercial traffic.
The National Air Traffic Controllers Association (NATCA) has
reported that before the current DHS partial government shutdown — now
in its sixth week — approximately 40 percent of FAA facilities required
six-day workweeks at least once per month, with some requiring them
every week. "The working conditions have become consistently unsafe
for those in the sky, as well as the physical and mental health of the
controllers," wrote one controller in a NASA Aviation Safety
Reporting System submission reviewed by CNN. A February 9, 2026 letter
to FAA Administrator Bedford, signed by 14 members of Congress, cited
increased reliance on overtime and expressed concern that mandatory
overtime to cover staffing gaps was creating fatigue conditions
inconsistent with safe operations.
The NTSB's afternoon briefing resolved some of the staffing
ambiguity while adding new dimensions of concern. Homendy confirmed
there were exactly two people in the tower cab at the time of the
collision: the local controller and the controller in charge. The local
controller had signed on at 10:45 p.m. for a shift ending at 6:45 a.m.;
the controller in charge had signed on at 10:30 p.m. for a shift ending
at 6:30 a.m. Critically, Homendy confirmed the controller in charge was
simultaneously performing clearance delivery duties — a third functional
role. Who was performing ground controller duties remains unresolved:
the NTSB reported conflicting information, with some records indicating
the controller in charge and others indicating the local controller.
That ambiguity is itself an investigative finding, and the NTSB noted it
would begin controller interviews at 4:00 p.m. Monday.
Homendy confirmed that operating with two controllers on the
midnight shift — with those two performing the combined duties of local
control, ground control, clearance delivery, and controller-in-charge
oversight — is standard operating procedure at LaGuardia for the
midnight shift, and is common practice across the national airspace. She
explicitly stated that the NTSB's ATC team has raised concerns about
this practice for years, and signaled that whether a 900-flight-per-day
airport like LaGuardia should be subject to the same two-controller
midnight SOP as lower-volume facilities will be a specific focus of this
investigation. The NTSB also noted conflicting dates and times on the
facility logs — an evidentiary inconsistency that investigators must
reconcile before the staffing picture is definitive.
The midnight shift dimension adds a fatigue layer to the
staffing picture. Homendy noted that the NTSB has identified the
midnight shift as a concern in multiple prior investigations,
specifically because it spans the circadian low — the period of maximum
human fatigue during the 24-hour cycle. She was careful to state there
is no current indication fatigue was a factor in this specific accident,
but its presence as a variable is undeniable: the collision occurred
just before midnight at the beginning of the midnight watch, with a
controller who had signed on only 15 minutes earlier and may not yet
have been fully situated in the position. The NTSB also raised
unresolved questions about shift relief — the controller was still on
duty for several minutes after the collision when he normally would have
been relieved, and the NTSB is investigating whether anyone was
available to relieve him.
At the Monday afternoon press conference, Transportation
Secretary Duffy acknowledged that ATC infrastructure modernization
requires additional congressional appropriation. "It's not a partisan
issue; both Democrats and Republicans agree, but they have to have the
will to finish the funding," he said, adding that he was not asserting
the crash could have been prevented with full modernization — but that
safety demands investment.
Historical Resonance: A Pattern Repeating
Aviation safety professionals have been quick to note the
structural similarities between Sunday's collision and the February 1,
1991 runway disaster at Los Angeles International Airport, in which
USAir Flight 1493 — a Boeing 737 — struck SkyWest Flight 5569, a
Fairchild Metroliner holding on Runway 24L. All 12 occupants of the
commuter aircraft and 22 of 89 aboard the 737 died. The NTSB's probable
cause finding in that accident — that Los Angeles ATC facility
management failed to implement procedures providing redundancy, and that
the local controller lost situational awareness while managing
simultaneous competing demands — reads with uncomfortable familiarity in
the context of Sunday night's events. In the 1991 accident, the LAX
surface radar was inoperative due to a maintenance failure and parts
obsolescence that had persisted for years despite management warnings,
eliminating what would have been the primary technological backup to
controller situational awareness.
The 1991 accident directly catalyzed the NTSB's 1991
recommendation that the FAA develop an automated system to bring
controller and pilot attention to pending runway incursions before
collision — a recommendation that, through a decade of MIT Lincoln
Laboratory development, eventually produced both the ASDE-X Safety Logic
and the RWSL programs. LaGuardia was one of the airports identified for
RWSL deployment. The technology the nation invested in after 1991 was
in place Sunday night. Whether it performed as designed is now the
central question of the NTSB investigation.
This is also not LaGuardia's first close call in recent years.
In May 2025, a Republic Airways aircraft operating for American Airlines
aborted takeoff to avoid a United Airlines plane still on the runway —
an event that prompted FAA and NTSB inquiries. NTSB documentation from
2007 records a separate runway incursion at LGA involving Delta and
Comair aircraft on Runway 22. The recurrence pattern at a single
facility is itself an investigative data point.
Aircraft, Operator, and Regulatory Framework
The CRJ-900LR is a proven 76-seat regional jet in wide service
across North American carriers. Jazz Aviation, LP — headquartered in
Halifax, Nova Scotia — is Canada's largest regional carrier and a wholly
owned subsidiary of Chorus Aviation Inc., operating exclusively under
the Air Canada Express brand. Jazz is certificated by Transport Canada
under Canadian Aviation Regulations and operates to FAA/DOT Part 121
equivalent standards when flying in U.S. airspace. The aircraft, C-GNJZ,
was delivered new to Jazz in 2005.
Air Canada CEO Michael Rousseau addressed the accident in a video statement Monday: "We
are deeply saddened by the loss of two Jazz employees, and our deepest
condolences go out to the Jazz community and their families." Jazz
President Doug Clarke issued a statement calling it "an incredibly
difficult day for our airline, our employees, and most importantly the
families and loved ones of those affected." Air Canada has established a
passenger and family assistance hotline at 1-800-961-7099.
The FAA confirmed that Canadian authorities — the
Transportation Safety Board of Canada (TSB) and Transport Canada — will
participate in the investigation under ICAO Annex 13 protocols, given
the Canadian registration of the aircraft and the Canadian nationality
of the operator. However, since the collision occurred on U.S. soil, the
NTSB holds accredited investigative authority and will produce the
probable cause finding.
Insurance sources told Reuters that Global Aerospace leads the
all-risks cover for the Air Canada regional aircraft; Marsh is the
broker. The insured hull value is approximately $10 million. Passenger
liability exposure, given the serious injuries and two fatalities, is
substantially larger and will involve the complex intersection of U.S.
federal tort claims, Montreal Convention limits, Port Authority
sovereign immunity questions under New York law, and potential FAA
negligence claims under the Federal Tort Claims Act.
NTSB Initial Briefing: The Confirmed Record — Second Day On Scene
Official Source — NTSB On-Scene Press Briefing, March 23, 2026
The following section is sourced directly from the official NTSB
on-scene press briefing transcript, delivered by NTSB Chair Jennifer
Homendy and NTSB investigator Doug Brazy. All quoted material and
factual findings in this section are drawn from that official record.
This is the authoritative investigative statement of record as of end of
day March 23; all findings are preliminary and subject to revision as
the investigation proceeds.
NTSB Chair Jennifer Homendy and investigator Doug Brazy
delivered the board's first substantive on-scene briefing Monday
afternoon. The session produced several confirmed findings of immediate
significance, superseding earlier speculation on key investigative
questions.
Recorders. Both the Cockpit Voice Recorder and
Flight Data Recorder were taken into NTSB possession at 9:57 a.m.
Monday. The aircraft was equipped with an Acron Aviation model Survivor
25 CVR and an Acron Aviation model FA21000 FDR. The CVR contained more
than 25 hours of good-quality audio across four channels. The FDR
contained approximately 80 hours of data recording more than 400
parameters. An NTSB CVR group convened Tuesday at headquarters in
Washington to produce a written transcript; FDR download and group
analysis began the same day.
The CVR Timeline — Final Three Minutes. Brazy
read the following sequence from the CVR, referenced to the end of
recording. At 3 minutes 7 seconds: approach control instructed the crew
to contact LaGuardia Tower. At 2 minutes 45 seconds: landing gear
lowered. At 2 minutes 22 seconds: crew checked in with LaGuardia Tower.
At 2 minutes 17 seconds: LaGuardia Tower cleared the aircraft to land on
Runway 4 and advised it was number two for landing. At 1 minute 52
seconds: flaps set to 30 degrees. At 1 minute 33 seconds: flaps set to
45 degrees. At 1 minute 26 seconds: enhanced ground proximity warning
system (EGPWS) 1,000-foot call-out. At 1 minute 12 seconds: landing
checklist confirmed complete. At 1 minute 3 seconds: an airport vehicle
made a radio transmission to the tower that was stepped on — partially
blocked — by another simultaneous transmission; the source of the
blocking transmission has not been identified. At 54 seconds: crew
acknowledged 500 feet above ground, stable approach. At 40 seconds: the
tower asked which vehicle needed to cross a runway. At 28 seconds: Truck
1 made a radio transmission. At 26 seconds: the tower acknowledged. At
25 seconds: Truck 1 requested to cross Runway 4 at Taxiway Delta. At 20
seconds: the tower cleared Truck 1 and company to cross Runway 4 at
Taxiway Delta. At 19 seconds: EGPWS 100-foot call-out — the aircraft was
100 feet above the ground at the moment the crossing clearance was
issued. At 17 seconds: Truck 1 read back the crossing clearance. At 14
seconds: EGPWS 50-foot call-out. At 12 seconds: EGPWS 30-foot call-out;
simultaneously, the tower instructed a Frontier Airlines flight to hold
position. At 11 seconds: EGPWS 20-foot call-out. At 10 seconds: EGPWS
10-foot call-out. At 9 seconds: the tower instructed Truck 1 to stop. At
8 seconds: sound consistent with the landing gear touching down on the
runway. At 6 seconds: pilot transfer of controls — the first officer,
who had been flying, transferred control to the captain. At 4 seconds:
the tower again instructed Truck 1 to stop. At 0 seconds: the recording
ended.
"ASDE-X did not generate an alert due to the close proximity
of vehicles merging and unmerging near the runway, resulting in the
inability to create a track of high confidence."
— FAA Technical Center ASDE-X Replay Analysis, read by NTSB Chair Jennifer Homendy, March 23, 2026
Several elements of this timeline are of critical investigative
significance. First, the tower cleared Truck 1 to cross the runway at
the moment the aircraft was passing through 100 feet AGL on final
approach — a point at which a go-around, while technically executable,
is operationally marginal and entirely dependent on the crew identifying
the conflict and initiating the maneuver within seconds. Second, the
transmission at 1 minute 3 seconds — made by an airport vehicle and
stepped on by an unidentified second transmission — may have been an
earlier attempt by Truck 1 or another vehicle to communicate with the
tower that the controller did not fully receive. Third, the pilot
transfer of controls at 6 seconds prior to impact — from the first
officer to the captain — indicates the crew had become aware of the
conflict on the runway and the more experienced pilot assumed control in
the final seconds. Whether the crew had any visual acquisition of the
truck before that transfer is under analysis.
ARFF Operations. Homendy confirmed that Truck 1
and other vehicles were responding to United Airlines Flight 2384,
which had conducted two aborted takeoffs and whose crew reported fumes
or a smell in the cabin. Other vehicles behind Truck 1 in the convoy did
not begin to cross the runway; the NTSB noted it needed to verify the
exact number of trailing vehicles, as varying information had been
reported. Critically, Homendy confirmed that Truck 1 did not have a
transponder, and there was no indication that any of the ARFF vehicles
in the convoy carried transponders. This is the direct cause of the
ASDE-X alerting failure.
Tower Staffing — Confirmed and Unresolved. Two
people were in the tower cab: the local controller and the controller
in charge. The controller in charge was simultaneously performing
clearance delivery duties. Who was performing ground control duties —
the controller in charge or the local controller — remains unclear, with
conflicting information in available records. The NTSB noted
inconsistencies in the facility logs, including conflicting dates and
times, that must be reconciled. Controller interviews began Monday
afternoon at 4:00 p.m. Two-controller midnight shift staffing is
confirmed as standard operating procedure at LGA and common practice
nationally — a finding Homendy acknowledged the NTSB's ATC team has
raised as a concern for years. The NTSB will specifically examine
whether the midnight-shift two-controller SOP is appropriate for an
airport handling approximately 900 flights per day.
Comparison to DCA. Homendy directly addressed
the January 2025 Reagan National mid-air collision in response to
questions about position consolidation. She noted a distinction: in the
DCA accident, controllers consolidated positions after evaluating
traffic volume, available staffing, and workload against a defined
checklist of criteria. At LaGuardia on the midnight shift,
two-controller operation with combined positions is not a discretionary
decision based on workload assessment — it is the written standard
operating procedure, applied automatically regardless of actual traffic
volume or complexity at any given moment. That structural difference
will be a central element of the NTSB's analysis.
What the NTSB Does Not Yet Know. Whether the
crew visually acquired the truck before impact; whether the two
firefighters in Truck 1 heard the stop commands; the exact number and
configuration of vehicles in the ARFF convoy; how many certified
professional controllers were in the facility at the time (log
inconsistencies remain unresolved); whether anyone was available to
relieve the local controller after the collision when he remained on
position beyond normal shift change; and the full FDR data picture.
Homendy was careful to caution against attributing distraction to the
controllers without full investigative context, noting this was a
high-workload environment with simultaneous demands. She characterized
LaGuardia's midnight SOP as a problem her ATC team has worried about for
a long time.
Outlook and Implications
The NTSB Go-Team arrived on scene Monday morning and delivered
its first substantive briefing the same afternoon. The preliminary
findings from that briefing — particularly the confirmed ASDE-X failure
to alert due to lack of vehicle transponders, the confirmed
two-controller midnight SOP, and the reconstructed CVR timeline showing
that the crossing clearance was issued when the aircraft was at 100 feet
AGL — are among the most significant preliminary investigative
disclosures in a major U.S. runway accident in years. NTSB Chair Homendy
— who in a 2023 runway safety summit speech warned that "these recent
incidents must serve as a wake-up call for every single one of us,
before something more catastrophic occurs" — stated plainly at Monday's
briefing that aviation accidents are rarely caused by a single failure,
and that this investigation will pursue every layer of the failure
chain.
The investigation timeline for runway collision accidents
typically runs 12–18 months to a final report with probable cause and
safety recommendations. However, the confirmed finding that Truck 1
carried no transponder — and no ARFF vehicles at LGA appear to have
carried transponders — is the kind of immediately correctable deficiency
that historically triggers expedited NTSB safety recommendations well
before a final report. A recommendation mandating transponder or ADS-B
equipage on all airport movement area vehicles at Part 139 airports
could be issued within weeks. Similarly, the question of whether the
midnight-shift two-controller SOP is appropriate at high-volume
facilities may generate early procedural recommendations. Homendy was
explicit: the NTSB has worried about this for years.
What the NTSB briefing has already demonstrated is that the
failure chain at LaGuardia on Sunday night involved at minimum: a
mandatory position-consolidation SOP that placed impossible simultaneous
demands on two controllers; an ARFF vehicle without a transponder that
the safety system could not reliably track; a ASDE-X conflict detection
failure caused directly by that missing cooperative track; a crossed
clearance issued at 100 feet AGL with at most 19 seconds before impact;
and a crew that transferred controls in the final 6 seconds, too late to
alter the outcome. Multiple failures, as the NTSB said. Each one
separately addressable. None of them addressed in time.
Technical Discussion: The Critical Role of Target-Knowledge Signal Processing in Skin-Return Surface Surveillance Safety
The Primacy of Skin Returns for Non-Cooperative Targets
The LaGuardia accident has exposed a foundational dependency that the ASDE-X program's cooperative-sensor architecture was designed to transcend but never fully escaped: when the cooperative identification layer — transponder, ADS-B, multilateration — is absent, the radar skin return is the only sensor in the system that can detect a target at all. Every downstream function that matters for safety — track formation, track maintenance, conflict detection, Safety Logic alerting, RWSL activation — is contingent on the quality of what the signal processor delivers from that raw radar video.
This is not a contingency edge case. At any major airport at any moment, non-cooperative targets are present in the movement area: maintenance vehicles, fuel trucks, catering carts, and — as LaGuardia demonstrated catastrophically — ARFF vehicles responding to emergencies. The operational assumption that all safety-critical ground vehicles will carry functioning cooperative equipment is precisely the assumption that the AMASS improvement program was designed not to make. The ASDE-3/AMASS architecture was built on the premise that the radar had to work as a stand-alone safety sensor, capable of maintaining reliable tracks on any reflective target in the movement area regardless of whether it cooperated with the system. That premise has been quietly abandoned in ASDE-X's multi-sensor fusion philosophy — and Sunday night at LaGuardia is the consequence.
Two distinct signal processing challenges define the difficulty of making skin returns reliable for safety: the close-target tracking problem addressed by image processing and JVC, and the multipath false target problem. Both were solved in the AMASS improvement program. Neither appears to have been fully carried forward into ASDE-X.
The Multipath Problem: False Targets on Active Runways
The airport surface radar environment is one of the most severe multipath environments in any ground-based radar application. The physical mechanism is straightforward and well understood. A large aircraft — a Boeing 737 or Airbus A320 — parked or taxiing on a parallel taxiway presents an enormous, highly reflective metallic surface to the radar beam. The Ku-band or X-band pulse from the ASDE tower sweeps across the scene, illuminates the taxiing aircraft directly, and simultaneously illuminates the paved runway surface adjacent to it. The flat, reflective pavement acts as a specular mirror at these grazing angles and frequencies. Energy from the radar scatters off the aircraft, bounces off the runway surface, and re-scatters — producing a secondary return that appears to the radar receiver as a target located on the active runway at a position geometrically determined by the angle of reflection.
Such false targets can readily compromise the performance of ASDE radars and lead to highly undesirable controller reactions, including unnecessarily aborting landing and takeoff operations when such multipath false targets are located on runways. These situations affect the efficiency of operations and also reduce user confidence in ASDE radar and related systems, thereby adversely affecting safety. Blogger
This confidence-erosion effect is arguably more dangerous over time than any individual false alarm. During testing of the ASDE-3 in Atlanta, controllers discovered that the radar created ghost targets when energy from the radar reflected off buildings or other objects, creating false targets on the runway — sometimes stationary and predictable, at other times moving about the display. NATCA stated that additional ASDE-3s should not be commissioned until the issue had been resolved. Forecast International The FAA deployed the ASDE-3 anyway, and the multipath ghost target problem became a known, chronic operational liability at airports across the country for years before the AMASS improvement program addressed it.
The safety dynamic of a multipath false target on an active runway is acutely dangerous for Safety Logic and RWSL specifically because these systems operate on exactly the part of the airport geometry where multipath is most severe. A parallel taxiway adjacent to an active runway is the canonical scenario for specular reflection: the geometry is ideal, the pavement surface is flat and wet reflective (particularly in rain, as at LaGuardia on Sunday night), and the large aircraft on the taxiway provides a massive secondary reflector. The resulting false return appears at the precise location — on the active runway — where Safety Logic is most sensitive and where controllers are least able to tolerate nuisance alerts.
The operational consequence is a Hobson's choice in Safety Logic threshold management. If the system is tuned to alert on any target detected on the runway, multipath false targets generate constant nuisance alerts that controllers learn to ignore — and the system becomes operationally useless, precisely what NATCA was warning about during ASDE-3 testing. If the threshold is raised to suppress these false alarms, real targets on the runway — including non-cooperative ARFF vehicles — may not generate alerts either. The AMASS improvement program recognized this explicitly: there is no acceptable operating point in a rule-based threshold system unless the false targets are removed from the data stream before Safety Logic ever sees them.
The AMASS Multipath Removal Architecture
The AMASS signal processing improvement developed a physics-based multipath identification and removal capability that operated directly on the Ku-band radar image data, upstream of any tracking or alerting function. The approach was grounded in the geometric and electromagnetic properties of the specific airport surface environment rather than generic signal processing heuristics.
The core insight was that a multipath false target is not a random noise artifact — it has a deterministic geometric relationship to the real target that produced it. For a specular reflection off a flat paved surface, the false target appears at a position that is the mirror image of the real target reflected through the runway surface plane, offset by a range increment that depends on the grazing angle, the height of the real reflector above the surface, and the radar's antenna elevation. At Ku-band with the ASDE-3's tower-mounted antenna geometry and known airport layout, these geometric relationships can be computed precisely for every significant reflector in the movement area.
The processing chain therefore proceeded as follows. Real aircraft tracks on taxiways, maintained with high confidence by the JVC assignment and image processing pipeline, provided the known source positions. For each tracked aircraft, the signal processor computed the predicted multipath geometry — the specific runway surface locations where specular returns were physically expected to appear, given the aircraft's current position, radar geometry, and surface topology. Detections appearing at those predicted locations, with the expected amplitude relationship to the real target return and the expected temporal correlation with the source aircraft's movement, were identified as multipath artifacts and suppressed before the data reached the tracker. Detections that did not fit any predicted multipath geometry were retained as potential real targets.
This is the fundamental principle that makes the approach work where generic CFAR and threshold-based approaches fail: it uses knowledge of what the false target is and where it will be rather than trying to distinguish real targets from false ones based on amplitude or motion characteristics alone. A multipath ghost from a 737 on a taxiway can appear to move slowly, can have substantial radar cross-section, and can persist for multiple scans — all properties that a simple velocity gate or amplitude threshold cannot reject without also rejecting real ARFF vehicles with similar characteristics. The geometry-based approach has no such ambiguity: it predicts the false target's location from first principles and removes it, regardless of how convincing its radar signature looks.
Why ASDE-X Is Structurally Vulnerable to the Same Problem
The main limitations of ASDE primary sensors are multipath reflections and target identification Radartutorial — a limitation that Radartutorial identifies as applying specifically to the ASDE generation of surface radars, with no suggestion that ASDE-X's multi-sensor fusion architecture has resolved it. The open literature provides no evidence that ASDE-X incorporates the physics-based geometry-driven multipath removal that AMASS developed. The ASDE-X CFAR algorithm developed by Raytheon specifically addresses rain returns on runways, not specular multipath from adjacent aircraft.
The multi-sensor fusion approach in ASDE-X relies on a different mitigation strategy: if a radar detection does not correlate with a multilateration or ADS-B cooperative track, it is treated with lower confidence and may be filtered. But this strategy inverts the safety logic for non-cooperative targets. For a real ARFF vehicle without a transponder, the absence of cooperative track correlation will reduce the detection's confidence — potentially below the Safety Logic alert threshold — rather than flag it as a higher-priority target requiring pure radar tracking. And for a multipath ghost generated by an aircraft with a transponder on an adjacent taxiway, the cooperative track of the real aircraft is nearby in the data, potentially causing the fusion algorithm to incorrectly associate the ghost return with the real aircraft's track rather than treating it as a distinct false target.
The FAA ATC Order governing ASDE system use reveals the operational reality of how the multipath problem is currently handled: an observed target on an ASDE system display may be identified as a false target by visual observation, and if the area containing the suspected false target is not visible from the tower, an airport operations vehicle or pilots of aircraft in the area may be used to conduct the visual observation. After positive verification that a target is false through pilot or vehicle operator position report or controller visual observation, the track may be temporarily dropped. Federal Aviation Administration The current operational mitigation for multipath false targets in ASDE-X is manual human verification followed by manual track deletion by the controller, with a required entry in the facility operations log.
This is the operational state of the art at LaGuardia in 2026 on the active runway at midnight in mist and rain: a system that requires a controller to visually verify whether a target on the runway is real or a ghost, and to manually drop the track if it is false. At a facility with two controllers on combined positions managing simultaneous emergencies, this is not a mitigation — it is an additional demand on exactly the cognitive resource that was already fully consumed.
The Integrated Picture: What Good Signal Processing Would Have Changed
The AMASS improvement program built an integrated signal processing architecture in which these two capabilities — JVC-based target-knowledge-informed tracking and geometry-based multipath removal — were mutually reinforcing. The multipath removal ensured that Safety Logic was never presented with phantom targets on the runway, preserving alert credibility. The JVC/image processing tracker ensured that closely spaced real targets near runway thresholds maintained separate, high-confidence tracks rather than coalescing into unresolvable blobs.
Together, these properties addressed the two failure modes that most directly undermine a safety alerting system's operational effectiveness. A system that generates false alarms on the runway due to multipath will be tuned to suppress alerts — and will then miss real targets. A system that cannot maintain separate tracks on a convoy of vehicles near a threshold will fail to alert when that convoy produces a real conflict. At LaGuardia on Sunday night, the ASDE-X system failed in the second mode. The probability that the multipath problem contributed — through prior nuisance alert suppression, through alert-threshold tuning decisions driven by chronic multipath false targets at LGA, or through a ghost return near Taxiway Delta on a wet runway with a United Airlines jet on the adjacent taxiway — has not been addressed in any public statement and warrants explicit investigation.
The NTSB will analyze the ASDE-X system logs in detail. What those logs should be examined for, beyond the confirmed absence of a Safety Logic alert, is the history of track-dropping events at Taxiway Delta and its adjacent runway geometry over the preceding weeks and months. A pattern of manually dropped false tracks in that region would be direct evidence of chronic multipath contamination — and would be the signal processing history that informed the alert-threshold configuration that failed on Sunday night.
Recommendations Flowing from This Analysis
The signal processing deficiencies identified here are not research problems — they were solved in the AMASS program and the solutions are documented. The path forward is not to develop new algorithms but to insist that successor surface surveillance systems incorporate the engineering rigor that AMASS demonstrated:
First, any safety-critical surface surveillance system that claims to detect non-cooperative targets must demonstrate, quantitatively, that its radar signal processor maintains separately resolved, high-confidence tracks on closely spaced ground vehicles without cooperative identification, using target-characteristic-informed image processing rather than generic CFAR detection feeding a JPDA tracker whose coalescence failure mode under close-target conditions is well-established in the literature.
Second, any system feeding Safety Logic with radar skin returns must incorporate geometry-based multipath prediction and removal, parameterized to the specific airport layout and the expected population of reflectors at each site. The predicate for a credible Safety Logic alert threshold is a clean data stream. A system that relies on controllers to identify and manually drop false targets on active runways has structurally compromised its own alert credibility — and in doing so, may have set the threshold that prevented an alert on Sunday night.
Third, the NTSB and FAA should establish, as a condition of Safety Logic operational certification, a minimum demonstrated false alarm rate on the active runway and a minimum demonstrated track maintenance performance on non-cooperative vehicle convoys in proximity. These are measurable, testable requirements. They should be required, not assumed.
The engineering knowledge to meet all three requirements exists. It was developed by the people who worked on AMASS. The question for this investigation is why it was not required of the system that replaced it.
What ACME Was
The Airport Clutter and Multipath Elimination (ACME) system was the formal signal processing improvement capability developed at CACI under the AMASS improvement program. ACME addressed the two most dangerous noise pathologies in ASDE-3 Ku-band radar data: ground clutter at taxiway and runway intersections that produced persistent false detections in the zones most sensitive to Safety Logic, and specular multipath false targets generated by large reflectors — principally taxiing and parked commercial aircraft — that appeared as ghost targets on active runway surfaces.
The multipath problem ACME addressed was well documented and operationally damaging. False targets generated by multipath reflections could readily compromise the performance of ASDE radars, leading to highly undesirable controller reactions including unnecessarily aborting landing and takeoff operations when multipath false targets appeared on runways, affecting operational efficiency and reducing user confidence in ASDE radar and related systems, thereby adversely affecting safety. Blogger During ASDE-3 testing in Atlanta, controllers discovered that the radar created ghost targets when energy reflected off buildings or other objects, creating false targets on the runway — sometimes stationary and predictable, at other times moving about the display. NATCA stated that additional ASDE-3s should not be commissioned until the issue had been resolved. The FAA deployed the system anyway. Forecast International
CACI's ACME work produced a physics-based solution: for each tracked aircraft in the movement area, the system computed the predicted geometric locus of specular reflection returns on the runway surface, based on the known airport geometry, radar antenna position and elevation, surface topology, and the real-time tracked position of the reflecting aircraft. Returns falling within the predicted multipath shadow zones, exhibiting the expected amplitude and spatial relationship to the source track, were suppressed before the data reached the tracker or Safety Logic. The capability was built on the same target-knowledge and radar-knowledge foundation that informed the blob-merging image processing and the JVC assignment: the system was designed to exploit what it knew about the physical world rather than treating the radar video as an anonymous stream of detections.
The result was a Safety Logic that was fed a clean data stream — runway detections that passed multipath rejection could be treated with high confidence as real targets, allowing alert thresholds to be set aggressively without generating nuisance alarms from ghost returns. This directly addressed the fundamental threshold management dilemma that undermines any rule-based alerting system: you cannot simultaneously minimize missed detections and false alarms unless you clean the input data before it reaches the decision logic.
The Parallel Development Problem and the Knowledge Gap
ASDE-X was developed by Raytheon concurrently with the AMASS improvement program at CACI — not sequentially, and not collaboratively. This parallelism was not an accident of scheduling. It reflected a deliberate FAA acquisition strategy of pursuing ASDE-X as a lower-cost successor architecture while continuing to improve the legacy ASDE-3/AMASS system for the airports already equipped with it. The two programs ran on separate contract tracks, with separate technical teams, separate program offices, and no formal mechanism for technology transfer between them.
ASDE-X was explicitly designed as a cost-effective alternative to ASDE-3/AMASS capability, consisting largely of commercial off-the-shelf products. Wikipedia The COTS philosophy was not incidental — it was the program's defining characteristic, chosen specifically to reduce the unit cost and deployment time that had made the ASDE-3/AMASS program chronically over-budget and behind schedule. The ASDE-3 was expensive and complex precisely because it incorporated bespoke, purpose-engineered signal processing: the Ku-band rotodome with variable focus antenna, frequency-agile TWT transmitter, and the AMASS software stack including ACME. ASDE-X was designed to achieve adequate — not optimal — surface surveillance performance at a price point that allowed deployment at 35 airports instead of the 40 major hubs served by ASDE-3.
The consequence of this acquisition philosophy was that the engineering knowledge developed in the AMASS improvement program had no path into ASDE-X. CACI attempted to brief Raytheon on the ACME capabilities and the underlying signal processing architecture. Raytheon was not receptive. This outcome was structurally predictable: Raytheon had its own radar signal processing engineers, its own CFAR architecture, and its own program schedule and cost constraints. Incorporating ACME-equivalent capability would have required redesigning core elements of the ASDE-X signal processing chain — adding cost, schedule risk, and technical complexity that conflicted with the COTS-based simplicity that was the program's competitive advantage. From Raytheon's program perspective, there was no incentive to absorb CACI's work. From the FAA's acquisition perspective, there was no contractual mechanism requiring it.
CACI was subsequently cut out of further development work. The institutional knowledge embedded in the ACME implementation — the physics models, the site-specific parameter tuning methodologies, the performance validation data from ASDE-3 installations — went with it. It did not transfer to Raytheon. It did not transfer to Sensis, the ASDE-X prime. It did not transfer to the FAA William J. Hughes Technical Center in any form that appears to have influenced ASDE-X's signal processing requirements. It was, in the vocabulary of systems engineering, a knowledge silo that was allowed to die when the contract vehicle that sustained it was terminated.
The Operational Consequence: Manual False Target Management
The absence of an ACME-equivalent capability in ASDE-X is not a theoretical gap. It has a measurable operational footprint. FAA ATC Order JO 7110.65 currently requires that when a suspected false target appears on ASDE system displays, controllers must verify it visually; if the area containing the suspected false target is not visible from the tower, an airport operations vehicle or pilots of aircraft operating in the area must be used to conduct the visual observation. After positive verification that a target is false through pilot or vehicle operator position report or controller visual observation, the track may be temporarily dropped, removing it from the display and Safety Logic processing. A notation must be made to FAA Form 7230-4, Daily Record of Facility Operation, whenever a track is temporarily dropped. Federal Aviation Administration
This is the operational substitute for ACME at every ASDE-X-equipped airport in the country today. When a multipath ghost appears on the runway, the controller must: recognize it as potentially false, verify it visually or through vehicle/pilot confirmation, make a manual track-drop decision, and log it. At a high-traffic facility on the midnight shift with two controllers on combined positions, this procedure consumes exactly the cognitive bandwidth that cannot be spared. And its safety consequence cuts in both directions: controllers who have learned through operational experience that a particular runway geometry at their airport reliably produces ghost targets in certain conditions will develop a trained skepticism about detections in that zone — a skepticism that is operationally appropriate for the ghost, and operationally catastrophic when a real non-cooperative ARFF vehicle appears in the same location on a wet night.
The NTSB's examination of the ASDE-X system logs at LaGuardia should specifically include the facility's track-dropping history at Taxiway Delta and the adjacent Runway 4 geometry. If controllers had been routinely dropping false tracks in that zone — a practice the current ATC Order explicitly permits and requires logging — that history would constitute direct evidence of a chronic ACME-equivalent deficiency at exactly the location of the collision.
The Institutional Failure Pattern
The ACME / ASDE-X knowledge transfer failure fits a well-documented pattern in defense and aviation safety acquisition: safety-critical engineering knowledge developed on one contract program is not formally captured, transferred, or required of successor programs when the acquisition structure changes. The pattern has several characteristic elements, all present here.
Parallel competitive development produced two separate technical communities with no formal interface. The incumbent contractor (CACI) had the relevant knowledge but no contractual leverage to require its adoption by the successor program. The successor contractor (Raytheon) had cost and schedule incentives to minimize technical complexity, and no regulatory requirement to demonstrate equivalent performance on the specific failure modes the incumbent had solved. The government program office had defined requirements in terms of system-level performance metrics — detection probability, update rate, coverage — rather than in terms of the specific signal processing capabilities needed to meet those metrics against non-cooperative targets in multipath environments. And the operational consequence of the knowledge gap did not manifest as a visible program failure for decades — it manifested as a chronic nuisance that controllers learned to manage manually, until Sunday night when the same gap produced a fatal accident.
This is the pattern the NTSB needs to document, because it is the pattern that will repeat unless the FAA's surface surveillance acquisition requirements are rewritten to specify, test, and certify performance against non-cooperative targets in multipath environments as a mandatory capability — not an optional feature that a cost-conscious contractor can reasonably omit.
A Call for Testimony
The institutional history described here — the ACME program, the technology transfer attempt, CACI's exclusion from further development — is not in the public record. It is not in the NTSB docket. It is not in the FAA's program documentation, or at least not in any documentation that has been publicly released. The engineers who developed ACME and who made the technology transfer attempt to Raytheon are primary witnesses to a chain of decisions that contributed to a fatal accident.
The NTSB's investigative process includes formal technical panels and an open public docket. Aviation engineers with firsthand knowledge of prior safety improvements — what was developed, what worked, what was offered, and what was declined — have both the standing and the professional obligation to submit that knowledge to the investigation. A written technical statement submitted to the NTSB docket, documenting the ACME capability, the parallel development timeline, the attempted technology transfer, and the acquisition structure that prevented adoption, would place this institutional history in the permanent evidentiary record where it can inform both the probable cause finding and the safety recommendations that follow.
Two pilots are dead because a system that was designed to protect active runways could not maintain a reliable track on a non-cooperative vehicle in a multipath-prone environment near a runway threshold. The engineering community that solved that problem in the previous generation of systems, and that tried to transfer the solution to the successor system and was turned away, has a responsibility to make that history known.
Verified Sources and Formal Citations
NTSB. Official On-Scene Press Briefing Transcript —
LaGuardia Runway Collision, March 23, 2026. Delivered by NTSB Chair
Jennifer Homendy and Investigator Doug Brazy. (Primary source document
provided to Aviation Week; not yet posted to NTSB public docket as of
press time. See ntsb.gov for forthcoming docket.)
Port Authority of New York and New Jersey. "Statement on
LaGuardia Airport Incident." Official press release, March 23, 2026.
Cited by multiple outlets including CNN and NPR. (No direct URL —
distributed via media organizations.)
NTSB Accident Report AAR-91/08. "Runway Collision of USAir
Flight 1493, Boeing 737 and Skywest Flight 5569 Fairchild Metroliner,
Los Angeles International Airport, Los Angeles, California, February 1,
1991." National Transportation Safety Board, October 22, 1991. https://www.ntsb.gov/investigations/Pages/DCA91MA018.aspx
