Current autopilot system is integrated with Flight Management System (FMS) for navigation such as LNAV and VNAV; both of them are based on preset parameters given by the pilot prior to departure (IRS alignment, total fuel/pax, waypoints/airways, flight level, etc) or in short, expert system

What are the restrictions for intelligent system autopilot in which the autopilot shall assist the pilot for making decision (or the autopilot decides per se)? For example, during flight under certain condition where some sensors are failed and flight control system reverts back to direct law (in case of FBW) and the intelligent system autopilot recognizes this case so it helps pilot to decide what to do or controls the aircraft thoroughly

  • 3
    $\begingroup$ If the computers were able to control the aircraft, then why would reverting to direct law (I think that's a strictly Airbus term, by the way, and not general to FBW systems) be necessary? The point of reverting to direct law is that the computers aren't able to safely control the aircraft in its current condition. $\endgroup$
    – user
    Commented Oct 10, 2018 at 9:58
  • $\begingroup$ @MichaelKjörling my mistake, sorry... yes, direct law refers to Airbus term for direct control from pilot to the flight controls; perhaps I was mixed up what I meant is should that case happened the autopilot does not disengage but rather keeps the aircraft in its current flight although it lost several sensors in order to fly normally (and other cases as well). Thanks $\endgroup$
    – dimas.w
    Commented Oct 10, 2018 at 10:31
  • $\begingroup$ You might be interested in Does engine failure shut off the auto-pilot? which, while not a duplicate, does discuss different autopilot modes on commercial aircraft and how the autopilot(s) can interact with various failures. $\endgroup$
    – user
    Commented Oct 10, 2018 at 10:48
  • 2
    $\begingroup$ What do you mean by "what are the restrictions"? Are you asking what legal restrictions there are on doing this? $\endgroup$ Commented Oct 10, 2018 at 11:19
  • $\begingroup$ @TannerSwett ... yes that's exactly what I mean, more to philosophical/regulatory aspect instead of technical $\endgroup$
    – dimas.w
    Commented Oct 10, 2018 at 15:56

2 Answers 2


Under the FAA there are not really regulations that limit the ability of an autopilot and shy of takeoff most modern systems are capable of performing most of a full flight with no pilot input (aside from route updates etc). Ultimately any aircraft system will have to get certified and this involves rigorous testing as well as demonstrations the system works under all stated cases.

There is however a bit of a logic issue in your statement:

For example, during flight under certain condition where some sensors are failed and flight control system reverts back to direct law (in case of FBW) and the intelligent system autopilot recognizes this case so it helps pilot to decide what to do or controls the aircraft thoroughly

The reason the autopilot disengages or disconnects is because, due to the sensor failure, it no longer has the information it needs to properly control the aircraft. No matter how smart the system is, without proper sensor input, positive control is not possible or would be logically unsound. This is even true for human pilots and was in part the issue that lead to the Air France 447 disaster.

There is a broader question here of autopilots potentially handling emergencies which is a bit different than a sensor failure (read pitot static failure). This is already in place in part on some aircraft and as time moves we see more and more of it. Aircraft are now capable of detecting a depressurization and automatically descending to a safe altitude; even new small GA planes have this feature. Autopilots also now have forms of automatic level mode which will bring the aircraft back to a straight and level attitude at the push of a button.

This also comes down in part to design philosophy. This is highly opinionated so I'll stay away from it on this site, but there is hard data to support both cases: in the event of an emergency, is a pilot or computer system better equipped to handle the situation?

  • 1
    $\begingroup$ One could make the case that, with a sufficiently robust ap, that if the ap doesn't have the necessary information to safely fly the plane then neither do the pilots $\endgroup$
    – TomMcW
    Commented Oct 10, 2018 at 19:41
  • 1
    $\begingroup$ @TomMcW yes I agree with you, should anything fails then the control reverts back to the pilot and at this present time (at least for now) we cannot rely on automated system entirely $\endgroup$
    – dimas.w
    Commented Oct 11, 2018 at 1:53

I'm not an expert, especially on the regulatory framework, but I can make a limited answer.

In some ways aircraft already are making incremental steps towards more automation and fail-proofing:

Airplanes already try to revert to basic functionality when something fails. Localizer beacon failure doesn't disengage the whole autopilot, and many autothrottles can work with one engine out. There may be some room for improvement in specific areas though - but bear in mind many of these sets of redundant sensors are already designed to have less than one complete failure per million flight hours.

Some of the problems don't lend themselves to an easy solution- if you're engaged in altitude hold and you lose your air data sensors, how do you measure altitude to maintain altitude? You could create a new sub-mode for level flight based only on the IRS, but this would be a lot of complexity compared to the alternative of hand-flying or just going to a basic pitch hold mode.

Sure, you could get around these dangerous scenarios by taking more authority in and reverting better to basic systems or modes when something fails, for example going into speed on pitch when both engines fail. Or in the case of Indonesia flight 8501 that you mentioned, forcing a backup flight computer to always be on during flight and somehow refusing to allow a dangerous roll to occur at the moment the autopilot disengages. However, in all these situations, you're trying outsmart the pilots and take decisions from them, which is... complex.

Some manufacturers give the pilot full authority, some limit the pilot only to what's known to be safe. While in my experience avionics engineers are happy to overthink things and try to idiot-proof their systems, these kinds of new systems come with a lot of difficulty in human factors testing, increased complexity, etc. and so changes are only made incrementally.

There is some regulatory work I'm aware of behind this. For example, planes are designed to be predictable and deterministic: the thought of untested scenarios or non-deterministic code keeps certification authorities up late at night. This limits development to maintainable, modular segments of flight and behavior. It's also easier to certify and train for incremental changes than a complete redesign, especially given how incredibly complex modern avionics are.

  • $\begingroup$ Thank you very much for your answer, so based on the description you wrote and by the FAA/CAA regulations it is possible to implement but we must also consider which field does the IS will take part especially on essential system like flight controls; this includes flight dynamics, electro/hydraulic/pneumatic actuators, sensors; outsmart the pilot by intelligent system is (IMHO) nearly impossible due to the fact that pilot must be the last line of defense should anything fails, perhaps the IS takes part as an advisory instead (if it is designed not to handle abnormal maneuver)? $\endgroup$
    – dimas.w
    Commented Oct 12, 2018 at 7:11

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .