Pilots get very upset if you suggest all they do is sit back, drink coffee, and watch the world fly by while the autopilot does everything for them. But there has been a push in the industry to think about a move to single pilot ops which raises questions over just how automated we should be aiming to be.
Let’s take a look at how automated we are now and whether we might be reaching a point where the human pilot might start to become redundant.
Support vs Supplant
Automation, up until now, has been there to help the pilots, not replace them. It does that by managing tasks which are capacity consuming or which require a high level of attention. Things like looking after the altitude or speed, or by taking on tasks which it is frankly just a bit better at. Like monitoring huge flows of data and checking for issues in it.
Modern aircraft have systems which enable them to do clever things like auto land. This isn’t for a scared pilot to use when it is windy out, but instead is used during low visibility operations to free up pilot capacity so they can monitor and make decisions on safety while the aircraft does the ‘basic’ stuff of flying.
The autopilot and auto-thrust system, along with the flight management systems, do a lot for the pilots—from maintaining trajectory and navigation, to providing protections against exceedances beyond the safe flight envelope. These systems help monitor fuel, engine performance and multiple other internal operations, and alert the pilot to issues and provide electronic checklists to help troubleshoot and resolve them.
There are systems for auto-starting the engines, and for attempting relights if they switch off in flight. There are automatic systems for preventing icing in flight, for transferring fuel… basically, all the guts and organs of the aircraft are automatically run and monitored for the pilots.
But while they have replaced the pilot in specific tasks, they have not been put in there to make the pilot redundant, but rather to support them and integrate with them, so that they can focus on the stuff only humans can currently do.
What automation do we have at the moment?
Well, beyond the obvious ones like the autopilot and auto throttle systems, there are some pretty innovative advances in modern aircraft. First up, fly-by-wire—something that has been seen on Airbus for decades now, across the majority of their modern models, but which is only really part of the wide body Boeing design (to compare the two major manufacturers).
Fly-by-wire replaces the old cables and levers mechanical methods of moving flight controls, and instead transmits pilot demands electrically to actuators. This has a lot of benefits. It reduces weight, maintenance and monitoring costs, and means a whole load more redundancy in the system. That’s not to say things don’t still go wrong. A recent Flexjets Legacy 450 had to return shortly after take-off in July after multiple flight control issues. Track that here.
But when it is all working right, fly-by-wire is a great thing and also fairly complex. Before it sends those signals, they are interpreted by flight computers and they do some clever stuff inside their brains. Things like looking after the flight envelope.
Airbus is the more automated of Boeing and Airbus, and this comes down to their design philosophy. Airbus wants to help mitigate against pilot error, whilst Boeing still says “the pilot has ultimate authority”. So, fly an Airbus and you will find a lot less feedback than Boeing, and get a lot less of a “real flying” feel (some pilots say it is less fun to fly), but the benefit is the level of protection built into the aircraft.
These systems prevent the aircraft going outside safe limits, and in Airbus they are automatic and are not overridable. In other words, the aircraft (if operating without any malfunctions) will not let you put it into an unsafe condition no matter how hard you try. Boeing on the other hand still will.
So we have automation in terms of handling specific systems, including doing the basic role of flying, and we have automation which looks after safety too.
More clever automation
The Airbus’ BTV system and their ROW/ROP functions are more examples of advanced automation. BTV, or brake to vacate, is a clever system which can calculate exactly how much deceleration is needed to meet a specific exit. Other aircraft tend to have auto brake, but this just provides a set deceleration rate, leaving the pilots responsible for making sure that is adequate for a specific exit (and for making sure they actually make the exit!)
ROW/ROP is an automated system which monitors whether the aircraft is going to be able to stop safely in the runway distance (left) available. If it thinks there is a problem, it can provide auto alerts to the crew, and on the ground will even set maximum braking and call for maximum reverse.
Then there is the automatic TCAS system – another feature unique to some Airbus aircraft at the moment. This is available on the A380 and A350 as standard, and it automatically flies a traffic avoidance maneuver for the pilots when a conflict is detected. All they have to do is let ATC know what is happening. In Boeing, and other aircraft, the TCAS monitoring and alerting system itself is automatic, but if a conflict is detected, the pilots still need to manually disconnect and fly the avoidance maneuver.
What automation can we expect in the not too distant future?
Airbus have added two new automations to the A350. The autopilot is able to recover automatically from a UAS situation—a major upset that takes the aircraft well beyond the normal flight parameters. Possibly more impressively, it can also fly an automatic emergency descent if pilots fail to respond.
But Airbus has not stopped there. A project entitled DragonFly is being tested out on the A350-1000 and if successful, this will enable automated emergency diversions in cruise, automatic landings and taxi assistance.
The automatic diversion could help in situations where crew are incapacitated by diverting to the nearest suitable and adequate airport. The ability to fly an automatic approach is also a huge advancement – while many an autopilot can currently do the trajectory management part of this, the pilot is still required for the configuration.
Advancements in communication and surveillance systems make this possible as the aircraft can automatically contact ATC, but the really clever innovation is in how the Airbus can “see” within its surroundings. Inspired by biomimicry, as seen in insects like dragonflies (hence the name), the systems can identify features in the environment and account for these, generating a new trajectory plan if any threats and hazards are spotted. It does this via a whole load of sensors, probes and cameras which input information to the internal computers enabling algorithms to analyze the trajectory.
Their ATTOL (automated taxi, takeoff and landing) project which includes their UpNext taxi assistance programme, will use systems which can identify and mitigate against hazards, while automatically controlling the aircraft trajectory on the ground. Since 2020, hundreds of flights have taken place to help them build the datasets for the algorithms. They have also carried out actual flight tests, with the ‘world’s first’ taking place in late June 2020.
Automation does not just lie within aircraft though. Projects to further automate other areas of the industry, including ATC, are underway. Machine learning and modeling can help optimize airspace usage, while managing separation, capacity and handling disruptions including weather which might lead to conflict or inefficiency.
What about military aircraft?
Well, the innovations in the military tend to be kept well under wraps, but an interesting development reported on quite recently took place at MIT.
If you’ve seen the impressive scene in Top Gun 2 where Tom Cruise manages to work out how to complete the impossible mission (I am trying hard not to give any details away for those who haven’t seen it) then this is an example of what current AI driven systems struggle with. Basically, the mission involves flying into a canyon so low to the ground that they evade any radar, and then have to climb up and out at an extreme angle.
That poses a bit of a problem for AI because it cannot think outside a ‘safe parameter box’ in the same way a human brain can. Automation, even the cleverest AI have to be programmed and this includes safe parameters, and in situations like the Top Gun one, it turns into a bit of a paradox because it cannot satisfy the conflicting goals of safety and stability, otherwise known as the ‘stabilize-avoid’ problem.
What does that mean? Well, it can’t both avoid the radar and not fly into the canyon wall. When it tries to solve one, it can’t solve the other, and vice versa. Or at least it couldn’t until MIT got at it. Earlier attempts at machine learning often resulted in time consuming trial and error where they would get one step closer to the solution for one bit, which would increase the challenge of the other, so then they would solve that, only to find now they have to re-address the other…
MIT reframed it as an optimisation propel instead. It gets very in-depth and mathematical, and for a better explanation, try reading this article, but what it means is a move to more advanced AI machine learning and automation, and what happens in the military often (admittedly after a lengthy time period) filters into commercial aviation as well.
Challenges to increased automation
A current challenge facing the aviation industry is GPS interference, in particular, spoofing. This is where aircraft GPS are sent false signals which corrupt the position and navigation inside the aircraft. GPS navigation is the primary navigation method nowadays, and not just en-route but for arrivals and approaches as well.
Degraded systems have wide ranging repercussions, from the ability to operate in reduced separation airspace, to an inability to fly RNAV/RNP approaches. Spoofing can also impact other systems like EGPWS which relies on the aircraft knowing its position (including its altitude) and being able to predict potential conflict with terrain, based on a database.
Another more hypothetical problem is ‘what if it goes wrong?” Currently, the human pilot retains the authority to remove automation and to take manual control from automated systems. While they cannot do this with all systems, they can with the primary ones. If the automation isn’t doing what it should, disconnect and revert to basics.
But as systems become more complex, the ability to do this decreases, and if we go as far as removing a pilot from the flight deck entirely, there is one safety level less already. Systems will need the ability to monitor the single pilot to prevent incapacitation, and if the pilot is ‘lost’, the ability to make decisions ‘outside the box’ might be lost as well.
While remote operators and decision makers are not a new thing – we see them in UAS operations already, and advances in BVLOS (beyond visual line of sight) operations are enabling this more and more—to integrate this when the size of the risk is so large (hundreds of passengers, hundreds of tonnes of aircraft) is a step people might not yet be ready for. Certainly, until the security of remote control is at a high enough level and issues with things like GPS interference are dealt with, this is not going to be a viable operation.
8 Responses
It appears to me that the human survival factor – the ability to work and earn a living – is being left completely out of the equation. As we automate more and more jobs letting AI do the work, we eliminate more and more jobs. To me this is a dangerous effect on society. The government loses tax money. How are all these people without jobs expected to survive?
I agree there always has to be a human overriding element. Pilots are trained to deal with problems of a feral flight system and take control of the aircraft while the fault is isolated.
A computer is only as good as the software is written.
I think most if not all of the things attributed to AI in this story are instead machine learning. When I put on the car GPS to give a final destination , its machine learning that works out the route and the guidance on the way. Same with airliners or military jets in their myriad software modes.
The story under sells Airbus FBW complexity. There are multiple modes and thinking they are infallible has led to many crashes and because its machine learning they will flip between modes and even to the mode where its all in the pilots controls. Even now the A320 series FBW doesnt cover the tail rudder, thats mechanical ( but changed only for the A321XLR)
The problem with eliminating human control altogether is that there is no longer anyone in control of that flight who has the ultimate investment in its safe outcome. Any mistakes that lead to catastrophe are made at such a distance that their effects on those responsible might be relatively insignificant. This isn’t an issue of blame and punishment, but rather of the missed opportunity and motivation to rectify any errors, besides the issue of passenger confidence.
Aircraft automation for general aviation aircraft is currently under development which will switch from 3 axis controls to a FADEC throttle for engine control and speed and a joy stick for navigation. The sensors and computer will manage keeping the plane upright and flying in trim. The pilot in ‘manual’ mode will simply keep the speed up and establish the course. Many more people will take up flying because the learning curve will be flattened. It is also believed that the general aviation flying experience with this gear will be a lot safer.
I don’t want a pilot unless I am the best pilot among humans
Having watched many TV documentaries about plane crashes and near misses, it is clear that in an emergency one pilot needs to devote 100% of attention to flying the aircraft, whist the other handles the Mayday communications, changes frequency, chooses between diversion airfields, talks with various controllers, etc. Between them they have to navigate and figure out what controls work and which do not, and to communicate with cabin staff and passengers. Navigators and flight engineers went long ago. And a single pilot can be taken ill, or sucked out through a window as happened with a BAC-111 once. Getting rid of the Co-pilot/First Officer would be a safety hazard
I haven’t flown often, and only on 747s, but I really do not like the idea of single pilot flights where many passengers are aboard.
I love aircraft, I was born just after WW2, my Dad was RAF, and my uncle worked on Vickers Armstrong aircraft, including the V-Bomber Valiant, Viscounts, Vanguards, VC10s, BAC111s and Concorde, mainly from Weybridge, although he also travelled a lot, as he troubleshot Flight Electronics and Simulators.