Aircraft automation – including that of commercial passenger aircraft – is becoming increasingly complex and, while the intention is to reduce crew workload and human error, pilots sometimes struggle to understand exactly what the automation (such as the autopilot) is doing ‘under the hood’, especially in off-nominal and emergency situations.
Coupled with this challenge is the fact that the aviation industry – including aircraft and avionics manufacturers – is pushing for Reduced Crew Operations (on long-haul commercial flights) and, in the longer term, Single Pilot Operations. This will inevitably lead to a higher degree of automation on the flight deck.
Coupled with this challenge is the fact that the aviation industry – including aircraft and avionics manufacturers – is pushing for Reduced Crew Operations (on long-haul commercial flights) and, in the longer term, Single Pilot Operations. This will inevitably lead to a higher degree of automation on the flight deck.
The SMARTAP R&I project aims to address the challenges described above by developing an ‘artificial pilot’ – based on Artificial Intelligence (AI) and Machine Learning (ML) – to assist the flight crew in high-workload situations. AI and ML are already being applied to many other domains – ranging from driverless cars to weather forecasting and cybersecurity – but their application to the aviation domain is still very limited due to the challenges associated with AI. Nevertheless, both the European Aviation Safety Authority (EASA) and Eurocontrol acknowledge the huge potential of AI and have recently issued reports explaining how AI could be applied to aviation and Air Traffic Management without compromising safety and security, while simultaneously keeping humans in the loop.
The first two years of SMARTAP focused on the development of ML models to automatically recover an aircraft from a loss of control scenario known as an aerodynamic stall. The models were developed using a combination of ML techniques including Reinforcement Learning, Behavioural Cloning and Deep Learning. The models were then trained by simulating various stall scenarios – including wings level (1G) stalls and turning stalls – in an A320neo flight simulation environment, with the assistance of expert human pilots. Finally, the models were tested by exposing them to over a hundred (unseen) stall scenarios at various altitudes and wind conditions. The results obtained have been very encouraging and demonstrate the suitability of AI and ML to loss of control situations.
The work described above was recently presented at the IEEE/AIAA 40th Digital Avionics Systems Conference through a research paper titled ‘Automated Aircraft Stall Recovery Using Reinforcement Learning and Supervised Learning Techniques’. The paper was presented by Mr Dheerendra Singh Tomar, one of the researchers working on the project.
The next phase of the SMARTAP project will focus on the development of ML algorithms to assist the flight crew during the final phases of flight – namely approach and landing – by detecting generic airfields from aerial imagery obtained by an electro-optical sensor mounted on the aircraft. This is expected to improve crew situation awareness, particularly in low visibility conditions.
SMARTAP is being led by Prof. Ing. David Zammit Mangion and Dr Ing. Jason Gauci from the Institute of Aerospace Technologies, in collaboration with Prof. Alexiei Dingli from the Department of AI (Faculty of ICT) and QuAero Ltd., a local aviation consultancy company.
SMARTAP (R&I-2018-010-T) is financed by Malta Council for Science & Technology, for and on behalf of the foundation for Science and Technology, through the FUSION: R&I Technology Development Programme.
