Industries / Transport and vehicles / Aerospace and defense
Avionics and flight control
This is the work of designing, building and proving the sensors, computers and software that measure how an aircraft is moving and move its controls, either to help the pilot or to fly the aircraft outright.
7
Milestones achieved, from 1914 to 2020.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1914
A gyroscope holds the wings
On 18 June 1914, at an airplane safety contest near Paris, Lawrence Sperry flew a Curtiss biplane fitted with a gyroscopic stabilizer, a set of spinning wheels that keep their direction in space, linked to the controls so the airplane corrects its own tilt. Sperry took his hands off the controls while his mechanic, Emil Cachin, walked about 7 feet (2 m) out along the wing, and the airplane stayed level. Sperry won the 50,000-franc first prize, by Aviation History magazine's account. The Smithsonian holds the winning unit.
Credit
Lawrence Sperry and the Sperry Gyroscope Co
1965
An airliner lands itself
On 10 June 1965 a British European Airways Hawker Siddeley Trident made what the Farnborough Air Sciences Trust museum calls the first automatic landing by a civil airliner in scheduled passenger service. The system could fly the approach, the flare (the last lift of the nose before the wheels touch), the touchdown and the roll-out.
Credit
Royal Aircraft Establishment, British European Airways, Hawker Siddeley and Smiths Industries
1972
Wires replace cables
On 25 May 1972 NASA pilot Gary Krier flew a modified F-8 that NASA calls "the first aircraft to fly completely dependent upon an electronic flight-control system." This is fly-by-wire: the pilot's stick sends electrical signals to a computer, which moves the control surfaces in place of cables and pushrods. The first computer was a spare Apollo flight-control computer, and later phases added three redundant computers and a software backup.
Credit
NASA Flight Research Center (now Armstrong Flight Research Center)
1988
Fly-by-wire carries passengers
The Airbus A320 was delivered to Air France on 28 March 1988, by Aviation Week's dates. Airbus calls it the first commercial aircraft with full digital fly-by-wire controls, and notes that fly-by-wire was first used on Concorde. It also brought flight envelope protection, software that stops the pilot from stalling the airplane, flying it too fast or banking it too steeply.
Credit
Airbus
1992
Rules for flight software
RTCA in the US and EUROCAE in Europe jointly write DO-178, the guidance regulators accept as the way to show airborne software is safe. It appeared in 1982, then as DO-178B in December 1992 and DO-178C in December 2011, which the FAA recognises. Software is graded from Level E up to Level A, where a fault could cause the loss of the aircraft. Level A requires MC/DC (modified condition/decision coverage): tests showing that each yes-or-no condition inside a decision can change the result on its own. A sister document, DO-254 (2000), does the same job for complex chips.
Credit
RTCA and EUROCAE
2019
One sensor grounds a fleet
On the Boeing 737 MAX, a flight-control program called MCAS could push the nose down again and again on data from a single angle-of-attack sensor, a vane on the nose that measures the angle between the wing and the oncoming air. By the FAA's review, one failed sensor drove the nose down in the Lion Air 610 crash (29 October 2018, 189 killed) and the Ethiopian Airlines 302 crash (10 March 2019, 157 killed), and the FAA grounded the type on 13 March 2019. The fix the FAA reviewed in November 2020 compares both sensors, acts once per event and leaves the pilots able to overpower MCAS with the control column. On 20 October 2026 the FAA certified the 737-10 with an enhanced angle-of-attack system, which US law then requires on every 737 MAX three years later.
LessonSoftware that moves the controls on its own must cross-check its sensors and must never be stronger than the pilot.
Credit
Boeing's redesign; the FAA's review
2020
Autoland for small aircraft
In May 2020 the FAA certified Garmin Autoland on the Piper M600, a single-engine turboprop. At the press of a button, or on its own, it picks an airport, flies there around terrain and weather, lands, brakes to a stop and tells air traffic control what it is doing. On 20 December 2025 it landed a Beechcraft King Air near Denver after what the operator reported as a rapid, uncommanded loss of cabin pressure, which Flightradar24 reported as the system's first use in a real emergency. Garmin says more than 2,500 aircraft carried the system by October 2026.
Credit
Garmin
Watch for
Autoland for helicopters
Skyryse says it will add emergency autoland to SkyOS, its fly-by-wire control system, and calls this "a first-ever to helicopter operations." Its release says helicopter autoland will follow certification of the Skyryse One helicopter, which the FAA had cleared in May 2025 to begin flight tests that count toward certification.
Achieved whenthe FAA certifies SkyOS on the Skyryse One and then approves the autoland function.
Claimed by
Skyryse, March 2026
Certified machine-learning avionics
In August 2025 the chief executive of Daedalean, a Swiss firm, said it was aiming to certify "the very first ML-enabled avionics" by the end of that year. The neural network, software trained on examples rather than written rule by rule, is trained, checked on separate data, then locked so it never learns in flight. Destinus agreed to buy Daedalean that month, and EUROCAE's standard for machine-learning avionics, ED-324, is still a draft with a target date of 31 December 2026. A search found no FAA or EASA approval since.
Achieved whenthe FAA or EASA approves an installed product whose function depends on a trained neural network.
Claimed by
Daedalean, August 2025
Catching every vane failure
Boeing says its enhanced angle-of-attack system adds five software monitors that check whether each nose vane behaves as expected, for example at takeoff and after a bird strike, and block or replace bad data. Boeing engineers say they searched several million 737 flight records for angle-of-attack sensor failures and that "every single one of those events would have been caught." The FAA certified the system on the 737-10, but the claim about past failures rests on Boeing's own analysis.
Achieved whena regulator, accident investigator or airline publishes its own check of the claim, or reports how the monitors perform in service.
Claimed by
Boeing, June 2026
Open
01
Open coupling checks for cFS
An open-source tool for NASA's core Flight System (cFS), free flight software used on NASA missions, that maps data coupling (one software part depending on data another writes) and control coupling (one part deciding when another runs) across the cFS core modules, then reports which of those links the existing functional tests actually exercise.
The briefCredit · unclaimed02
MC/DC for PX4 control laws
MC/DC measurement for the flight-control code of PX4, an open-source autopilot, plus new unit tests that bring one control library to full MC/DC, each test tied to a written requirement. Everything runs as unit tests on a desktop computer, and nothing flies.
The briefCredit · unclaimed03
Requirement trace for F Prime
A check that runs on every change to F Prime, the open flight software framework from NASA's Jet Propulsion Laboratory that ran the Ingenuity Mars helicopter. It reads the requirement table in each component's design document, finds the requirement tags that F Prime unit tests already carry, and reports requirements with no test, tests citing unknown requirements, and requirements marked for unit test that no test claims. DO-178C asks for this trace both ways, from requirements down to tests and back up.
The briefCredit · unclaimed
Open 01 · the brief
Open coupling checks for cFS
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source tool for NASA's core Flight System (cFS), free flight software used on NASA missions, that maps data coupling (one software part depending on data another writes) and control coupling (one part deciding when another runs) across the cFS core modules, then reports which of those links the existing functional tests actually exercise.
Why it is still open
DO-178C asks for evidence that tests exercised these couplings. Rapita, a tool vendor, writes that "there are no accepted standard criteria to identify coverage goals", and a 2015 SAE paper says this analysis has been "challenging in the past due to little tool support." Tools that do it today, such as VectorCAST/Coupling and Rapita's RapiCoupling, are commercial. A search found no open-source tool and no published coupling analysis of cFS.
Built with what exists
Clang/libclang or CodeQL to list the couplings from source; GCC's -finstrument-functions or LLVM XRay to record calls at run time; cFS built in its native Linux simulation; Python for the report; an AI coding agent to draft the parsers, with a person checking the output against a hand-built list.
First step
Build cFS natively on Linux, run the functional test app, and list by hand and by tool every call from the software bus module into other core modules.
Done looks like
A public report for one cFS core release listing every coupling between its core modules, each marked exercised or not by the functional test app, with the tool matching a hand-built list for one pair of modules with zero misses and running automatically on every commit.
Open 02 · the brief
MC/DC for PX4 control laws
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
MC/DC measurement for the flight-control code of PX4, an open-source autopilot, plus new unit tests that bring one control library to full MC/DC, each test tied to a written requirement. Everything runs as unit tests on a desktop computer, and nothing flies.
Why it is still open
PX4's coverage target counts lines and branches, not MC/DC, and a search of the PX4 and ArduPilot issue trackers found no MC/DC work. Free MC/DC measurement arrived in common compilers recently: GCC 14 (2024) added it, and NASA already runs it on cFS's core in its automated tests, which reported 99.68% of 8,032 condition outcomes covered in October 2026.
Built with what exists
GCC 14 or later (-fcondition-coverage, gcov --conditions) or Clang (-fcoverage-mcdc); PX4's existing GoogleTest unit tests; gcovr for reports; an AI coding agent to draft tests for a person to review.
First step
Build PX4's unit tests with -fcondition-coverage added for the rate-control library and list the uncovered conditions.
Done looks like
Published MC/DC percentages for PX4's rate controller, multicopter attitude controller and control allocator, with one of them reaching 100% of condition outcomes, measured automatically on every commit.
Open 03 · the brief
Requirement trace for F Prime
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A check that runs on every change to F Prime, the open flight software framework from NASA's Jet Propulsion Laboratory that ran the Ingenuity Mars helicopter. It reads the requirement table in each component's design document, finds the requirement tags that F Prime unit tests already carry, and reports requirements with no test, tests citing unknown requirements, and requirements marked for unit test that no test claims. DO-178C asks for this trace both ways, from requirements down to tests and back up.
Why it is still open
F Prime issue #1564 asked in July 2022 for "a comprehensive approach to requirements traceability", and it is still open. An October 2026 comment on it counts 419 requirements in 59 component design documents, 304 of which name unit test as the verification method. F Prime added an AI reviewer of requirement wording in October 2026, but nothing that checks the trace.
Built with what exists
Python; F Prime's design documents and REQUIREMENT macro; an open tracing tool such as OpenFastTrace or StrictDoc; GitHub Actions. The check itself stays plain code so its counts can be trusted.
First step
Write a parser for the design-document requirement tables and reproduce the count of 419.
Done looks like
A matrix covering every requirement in the framework's design documents, each marked traced or untraced and regenerated automatically, whose counts reproduce or correct the 419 and 304 in issue #1564.