A Conversation with CAE CTO / CPO Abha Dogra
EXCERPTED FROM THE BOOK, THE ROBOT IN THE SIMULATOR: ARTIFICIAL INTELLIGENCE IN AVIATION TRAINING, BY RICK ADAMS, FRAES.
Read the entire interview and case studies of CAE AI projects.
Available through Aviation Voices: https://aviationvoices.com/shop/
Rick Adams: I’ve been talking to a lot of people in the industry about AI. My general sense is that it’s early days. There’s a lot of exploring and experimentation going on. Maybe not a lot of hard implementation yet.
Abha Dogra: This is a very pertinent topic, Rick. I am part of many discussions with CIOs in the airlines world, be it in the supplier area within the aviation sector – Rockwell Collins, Boeing, Southwest Airlines, Emirates… and the reality is AI is actually disrupting every vertical. It’s just the pace at which it will be adopted.
I’m very proud to see that, even before my time, CAE was aware of this disruption, in a way, and was paving the way with its CAE RISE initiative. Maybe a little bit premature, in the sense for the rest of the market to understand what we are doing. But that gave us the runway to now be ready with our prime productionized solutions to help our customers work with regulators to evolve the standards as they are stepping into this area.
Rick Adams: You say RISE was premature when it was first started, but how have you matured it till now?
Abha Dogra: So if you look at the ‘flywheel,’ as I call it, of the training ecosystem –everything to do with the data at the time of training a pilot, all the way to see a pilot actually applying that training in the real world at the time of flying and doing the operations. This is the complete flywheel now where we sit at CAE today, given our huge network of 40-plus training centers.
Today we have almost 886 full-flight simulators integrated into our data lake. We are sitting on top of 2.5PB (petabytes) of training data, which relates to evaluating the insights at the time of training. To convert that into effectively help the pilots get better trained through instructors.
So we have a data lake where we have the most variety of aircraft types integrated to provide telemetry at the time of training a pilot that instructors can then evaluate while grading, and help the pilots be much more competent to really keep us safe.
On top of that, when we have that kind of flight training data, we also sometimes get data like LOSA (Line Operations Safety Assessments). As you know, LOSA flight operations data can, over time, triangulate together to see how the same pilot worked at the time of training, but also applied that training in real time during the flight operations. Obviously, we have much more scope to evolve in that ecosystem with the airlines because most of the flying data is with the airlines.
But that’s an opportunity to keep the world safe. For us to complete this flywheel and trace the competency of a pilot, both at the time of recurring trainings, ab initio trainings, keeping their certifications all the time, all the way to their actual operations competency. That’s one piece.
The second piece – apart from the telemetric data from the simulators, which is very procedural – the instructor is sitting behind the pilot and cannot always see the biometrics in terms of where is the gaze of the pilot. How is the pulse rate? It’s a dimension of the data that now we can apply.
So we have started to further augment our data lake with the biometrics data, obviously all under privacy compliance, approved by the pilots. We never do anything which is not in compliance.
Biometrics is another area where threat and error maneuvering is the biggest risk for the airlines. We typically train pilots in unusual situations, obviously; we subject them to a lot of scenarios where they maneuver, the weather conditions… Pilot gazes and pulse rate, these are things you cannot capture if you are not capturing the biometrics and augment the telemetric data of simulator to further harness the insight.
So that’s another area we are currently working on, to apply not just the simulator telemetric data but also the biometric data, to triangulate it, to derive the insights that the instructors can then share real time with the pilots to help develop their competencies.
Rick Adams: And what sort of biometrics have you found to be the most useful?
Abha Dogra: We studied 11 different biometrics. One we found the most optimal, which we are augmenting with Singapore Airlines, is around gaze tracking.
We are doing the same thing with the pulse rate with variables as to how does your pulse rate fluctuate under turbulent situations. That’s another kind of biometric we are starting to leverage, but more for the ab initio training.