Artificial Intelligence in Your Cockpit

1.    Artificial Intelligence in Your Cockpit

Artificial Intelligence is adopted widely in various areas including aviation domain. Artificial intelligence in the cockpit – cooperating with a pilot or even taking autonomous decisions– this opens new opportunities for functions which increase safety, reduce crew workload and fatigue, optimize flight trajectories for fuel efficiency in crowded airspace and open completely new business opportunities. 

Besides the benefits which AI will bring to the aviation segment, there are also challenges to be addressed. The AI based functions need to be evaluated not just from the it’s benefits point of view, but also need to be evaluated from:

  • Safety perspective – “How we can know the AI based function decisions are always safe?”
  • Ethical perspective – “How we know the AI based function is free of cultural/racial/gender… biases?”
  • Transparency perspective – “Can we determine based on what AI function took particular decision?”
  • Human perception perspective – “How pilots will accept AI functionality and what will be pilot’s expectations from AI functionality?”
  • Certification perspective – “How we can certify complex SW functionality which is Machine learning based?”

All these questions must be asked when we are thinking about bringing AI based functionality into the cockpit. Once we know the answers, then the technology revolution in the cockpit may begin.

1.1.     AFI-X Prototype Story

The beginning of the Artificial Instructor story (in 2015) was simple – “I wish to have someone experienced on board with me, when I’m flying solo.” In other words, the idea came from the community of pilots with low flight time – either new pilots with fresh license or pilots with long flight break.

The second step was also the simple one – Let’s build application, running in the cockpit, which monitors in real-time pilot performance and detects pilot mistakes which may develop into the real problem if not mitigated accordingly. Once the mistake is detected, the system will recommend corrective action. The system may observe pilot improvements over the time and change focus from major errors which may lead to unsafe situations towards smaller errors which are commented to improve pilot skills. In other words – the function will do what flight instructor is doing. And how to implement such functionality? The Artificial Intelligence definition (one of many) gives us the answer: “Artificial intelligence is the simulation of human intelligence processes by machines, especially computer systems.” (Definition is taken from techtarget.com).It means that simulation of human instructor intelligence will be Artificial Intelligence – Artificial Instructor. 

The idea was clear and desired AI functionality well defined, however the framework for development of AI based, real-time application, which assist to a pilot during flight – this framework was missing at that time – so we had to built the framework from scratch.

Note that today situation is better than our initial position in 2015. In 2020, EASA published the EASA AI Roadmap 1.0 and at the end of 2021 the First Usable Guidance for Level 1 Machine Learning Applications was published by EASA.  As we asked right questions back in the 2015, our framework we developed during the work on AFI-X prototype is well aligned with the EASA guidance and EASA roadmap. 

And mainly, our framework we developed is generic – we are able to tailor it development of various AI based applications and we are also able to evolve the framework from actual stage towards more autonomous AI/ML frameworks. The figure below is taken from EASA AI Roadmap 1.0 – it is clear that we are just at the beginning of the technology revolution in Aerospace.  

image

Figure 1: EASA published roadmap for AI applications. Taken from https://www.easa.europa.eu/downloads/109668/en, page 13.

The AFI-X journey was successful – the project moved from the initial idea through the AI algorithms concept definition towards simulator testing and finally was flown in real airplane for many hours assisting to pilots. This story helped us to validate the correctness of the framework and develop good AI based Level 1 application prototype.

image

Test at simulator…

image

… and real flight test.

The developed framework and processes can be re-applied and customized for development and certification of various AI based avionic applications – despite the selected algorithms and target functionality.

Are you interested about tips and tricks for a AI development and testing? Read more in next article AI Development – Tips & Tricks

Or would you like to know more about the AI development framework? Read more in the article Framework for AI Development and Certification