Airbus are innovating on automation – which is awesome/concerning depending on who you ask

If you ask me…

I say Airbus make clever airplanes (yes, Boeing do too, I love my 787).

Anyway, I can say that as someone who as flown a fair few of them (A330, A340, A380 and know a decent enough amount about the A320). The automation in current Airbus aircraft is awesome because, *used correctly, it ‘integrates’ with the pilot and assists and supports by reducing their workload, monitoring things it is hard for our slow little human brains to monitor that fast, and by offering layers of protection.

I am going to just repeat the *used correctly bit because this is really important. The automation is not there to replace the pilot, or the requirement for the pilot to be aware at all times of what the aircraft and its systems are doing. However, if you ask some, the most common negative argument against Airbus levels of automation is that it takes the pilots out of the loop.

Well, no, it doesn’t – unless the pilots don’t understand enough or become complacent and reliant on it (with the possible exception of the side stick, I’ll give you that).

The key point to Airbus’ philosophy is that the pilots must always ANC and use appropriate levels of automation. That might mean fully managed, automated levels which the pilots monitor. It might mean keeping the autopilot in but controlling the aircraft via it in basic modes. It might mean manual flying if it isn’t doing what you need it to. It always means knowing what you are asking the aircraft do, understanding what it is actually doing, and being in control of it.

**Conversations on automation complacency, reliance and the risk of mishandled/understood automation are a huge one though and I am not trying to have that right now, here. They should be had.

If you ask me about Dragonfly…

Dragonfly is the name for the new technologies Airbus are trialling through their ‘UpNext’ project. They are designed to ‘enhance pilot assistance’, and were recently demonstrated on Airbus’ A350-1000 Demonstrator aircraft.

In a nutshell, they include:

• Developments in flight safety
• Developments in operational efficiency
• Automatic landing, incapacity and taxi assistance

It is named ‘DragonFly’ because dragonflies are good at identifying features in the landscape and flying around them at high speeds (they are one of the fastest insects out there). Airbus are designing through biomimicry (copying nature). Also, dragonflies are a darn site more attractive than bluebottles, locusts or corn earthworm moths and the name has a much better ring to it.

Let’s start with the taxi

We see a lot of ground incidents – from aircraft collisions on the ground to runway incursions, and the primary feature of all of these are that they really could (should) be avoidable.

DragonFly assists with features such as audio alerts when it senses potential obstacles, assisted speed control and guidance about where the runway is with an assisted map. It can also carry out an auto landing and taxi to stand if it needs to.

This seems to be more of an evolution of existing systems, rather than an evolution.

We already see moving maps on aircraft, audio alerts regarding runways and camera systems to help indicate gear position. BTV allows aircraft to select and brake for a specific exit, and terrain warning systems are old news now. So systems which can alert better about where the runway is, whether wingtip clearance may be an issue and which can assist with speed control (good for taxiway excursions, winter ops, and brake temperatures) could be extremely beneficial.

But I know you’re here to hear about the flying bit

And I get it – because this is the technology which we all know this is likely a precursor technology to moves into single pilot ops. Whether it is there now, and whether this is a good thing (ever) is a whole discussion. I just want to see what this new tech is about.

The main feature of the technology seems to be that it can consider pilot incapacitation, and when detected can ‘generate a new flight trajectory’ (work out where to go and direct itself there), account for external factors such as flight zones, terrain and weather conditions, all the while letting ATC and the ops department know what its plan is. The talking with ATC and the operator is important because it allows a continuous flow of info, enabling it to adjust the flightpath and reselect more suitable options if required.

This could be extremely beneficial in supporting pilots during intensive phases like replanning diversions.

During a diversion, pilots need to weigh up a lot of options fast – from ‘can we actually get there?’ with the fuel and time available to ‘can we actually land, stop, go-around if we need’ which considers performance and weather limitations. In medical emergencies, we also have to weigh up the benefit of getting one person on the ground fast vs increasing the risk to all those onboard if the airport or environment is higher risk. This is not easy to do.

Support in analysis information, faster determination of options, and more support from the systems to initiate and look after the diversion do seem like a very beneficial safety advancement. However the system can also entirely take control in the event of full crew incapacitation. Airbus have not elaborated on this except to say it can detect it and takes over within a timeframe as required. Presumably, it is over-rideable if required.

This is quite revolutionary. The A350 currently has systems which can autonomously fly a rapid/emergency descent if pressurisation is lost, and which can automatically control the aircraft back into a safe state following an upset. However, these systems not only control the aircraft but have an element of decision making (data consideration and evaluation might be a more technical and less human term to use).

Let’s make a quick stop at MIT

Remember the epic scene in “Top Gun: Maverick” where they fly an impossible mission, zooming through a canyon so low the radars can’t spot them then make a death defying climb straight up and out? Well, while humans can do this, AI can’t. Or I should say couldn’t. I’ll get to that just now.

The issue for AI is something called the ‘stabilise-avoid problem’. Basically, the scenario represents a conflict between the most straightforward path towards achieving the purpose (in this case dropping something on a target) vs what need to happen to avoid crashing/radar detection. The opposing confections in the scenario make it impossible to resolve.

In the past, methods to resolve this have gone all ‘simple maths’ but fail to consider real world dynamics, making it rather unuseful. The MIT brains looked at it as more of a ‘constrained optimization problem’ – in other words, opened up the ‘goal region’ and through trial and error the AI learned and found a solution where it could reach and stabilise its goal.

OK, that explanation makes no sense because I don’t really understand it (link below to the article if you want to read it). This technology is being looked at for drones, but more of a military application. However, one of the big questions about the Airbus technology (for me) is how complex a decision can it make? And by that I mean when conditions becoming too challenging, but the requirement to divert is too pressing, how would it respond to this conflicts?

When the aircraft discovers its own ‘stabilise-avoid’ multi-objective problem where the dynamics are nonlinear and high-dimensional, what will it do?

Currently, the human ability to find alternative solutions or to sometimes break or bend one limitation or procedure in order to achieve a necessary outcome of safety remains far beyond an AIs.

The verdict

It isn’t really a verdict, I don’t have anywhere near the knowledge or understanding of the systems, but I do see the immense benefit of the technology which supports and assists pilots, particularly during taxi. The vast majority of incidents during the ground phase result from a loss of situational awareness by the crew, often down to distractions or confusions. This technology will help reduce the workload and support crew against these.

The in-flight technology also looks as if it can offer a huge amount of support during intensive phases, assisting with rapid data analysis and leading to potentially quicker and less bias laden decisions.

I recall one sim exercise I did where I was ‘single pilot’ attempting to address a fuel leak and weigh up the options of landing as soon as possible and probably losing an engine if not two (I had 4) during the landing phase, vs continue to try and minimise the leak. This is not what the DragonFly technology currently would assist with, but it could lead to better support in this type of situation.

However, an over reliance on automation is already something we see as an issue in the industry, so any system needs to be properly integrated with the pilots – they need to understand it and they need to retain the skills to handle situations without it.

Technology which hints of single pilot ops also raises questions – for me, these aren’t necessarily the questions over safety with only one human onboard (there is a whole lot to be considered there), but far more critical I think are questions about the mental well-being and confidence of the one pilot onboard, and the great loss of learning and development which we all do on both an instant and daily basis, and often a long term basis, when working with another colleague.

I also think that overall the likelihood of two pilot incapacitation is remote. Very remote. The likelihood of potential issues if we move too fast to single pilot ops is far less remote. Technology to replace a pilot therefore seems unnecessary, while technology to support pilots more is important, so I think this technology is amazing, will lead to a whole world of improvements in safety and efficiency, but I also (as I think many do) hope the right questions and conversations are had about what it might be allowed to lead to.

Read more, find your own opinion

There are various sources with some info about the new technologies:

Flight Global did a great write up on it in January

Airbus’ own releases on it

The Science Daily post on MIT studies

Part of the MIT study paper (if you want more crazy graphs and equations)