Artificial Intelligence in the Cockpit: A Comprehensive Review of an Existing Study on Artificial Intelligence in Flight Management Systems

By Sreeram Chittayil, Year 12 Student

Introduction

The aviation industry is entering a new era in adaptive and real-time flight data processing. Machine learning is one of the central technologies enabling this transformation, more broadly recognised as artificial intelligence (AI). Over the past decade, AI applications in aviation have expanded beyond traditional autopilot systems, extending into predictive maintenance, safety monitoring, and real-time route optimisation[1].

The Flight Management System (FMS) is at the centre of this transition; it is an onboard processing unit that provides navigation, fuel planning, and performance optimisation while the aircraft is in flight and on the ground. These systems were traditionally built on static instructions and set limits, with published navigation and performance databases that necessitated lengthy revision cycles whenever updates happened[2]. This rigidity limited their capacity to respond to real-time environmental and operational changes, resulting in inefficiency.

Recent advances, however, are testing these limits. Next-generation FMS models from Honeywell, Collins Aerospace, and Thales have begun to include data-driven frameworks that allow continuous input from meteorological services, air traffic systems, and onboard sensors[3][4]. Such systems process live data streams and utilise adaptive algorithms to assess a large number of variables at the same time, providing recommendations—or, in some circumstances, autonomous adjustments—in real time. This is a shift from past procedures, in which flights were virtually locked into plans set hours before departure.

This article will review and expand on a previously published study that examined AI integration in FMS using a purpose-built algorithm. The study created and tested an adaptive model capable of incorporating live flight data, and it displayed demonstrable accuracy in real-world circumstances[5]. When this study is seen in the context of the larger research environment, it becomes evident that such advances are not isolated experiments, but rather part of a rising industry-wide trend toward adaptive, AI-powered aviation.

The implications of this shift touch the most critical aspects of aeronautical engineering: reducing operational costs, lowering environmental impacts through optimised fuel use, improving passenger comfort through smoother route adjustments, and, most importantly, enhancing safety margins[6][7]. These outcomes demonstrate why AI-driven FMS are not just technological upgrades but are poised to redefine the future of aviation operations.

Background

Airlines have long sought methods to balance efficiency and safety, but the latest flight management systems (FMS) go beyond incremental gains. Research organisations like as MIT are actively working on predictive systems that assess infrequent failures in highly complex contexts, with new studies illustrating how computers may detect cascade disruptions well in advance[8]. These systems behave more like expert copilots who constantly reassess the circumstances, rather than navigators that stick to a fixed plan[2]. Every aircraft's systems communicate invisibly—engines vibrating at precise frequencies, hydraulics sustaining pressure, electronics drawing current. These signals, when regularly watched, can detect problems before they become failures, allowing for scheduled repairs and eliminating costly cancellations.

Real-time data integration is a key aspect of adaptive FMS systems. Unlike older systems, which relied on fixed forecasts or wide dispatch updates, modern systems continuously collect data from numerous sources, including weather satellites, online services, and onboard sensors. This enables them to detect turbulence or unstable air patterns early and recommend reroutes or altitude modifications before passengers suffer the consequences[2]. This ongoing revision is not restricted to weather; performance metrics like as engine health and aircraft weight distribution can also be considered, transforming flight planning into a dynamic, forward-looking process.

The implications and consequences of this transition are profound. Fuel, which accounts for almost 28.7% of airline operating costs and totalled more than $188 billion globally in 2019, is one of the most obvious beneficiaries[6]. Previously, personnel could only make broad changes for wind or weather, and studies such as Ramée's (2020) found that fixed modules frequently had inaccuracies of 30% or more[9]. Newer adaptive systems modify these decisions automatically, reducing inefficiencies by altering altitude, speed, or rerouting slightly when favourable conditions are identified. This type of optimisation directly lowers expenses and emissions.

Another consequence is increased safety margins. Pilots remain highly skilled decision-makers, but having a system that constantly scans for crash risks, performance limits, and unanticipated dangers gives them earlier warnings and more time to respond. In past fixed systems, pilots were frequently reacting to events after delays; adaptive FMS transforms this into a proactive process in which alerts and recommendations arrive minutes sooner, multiplying available options and boosting passenger comfort through smoother flights.

Finally, the progressive integration of artificial intelligence into avionics has far-reaching consequences in the industry. While it is still challenging to install full AI systems directly onto aeroplanes due to safety and certification problems, the benefits have already been demonstrated in ground-based systems and airline operations centers. Aerospace America (2023) reports that AI-assisted route planning technologies have already increased efficiency and dependability[7]. MIT's "Air-Guardian" project exemplifies the potential for AI copilots to supplement rather than replace human judgment[8]. Although adoption in the cockpit will be gradual, the trend is unavoidable given the enormous operational benefit AI provides for efficiency and safety.

When comparing fixed and adaptive architectures, the difference is obvious. Traditional FMS involved loading a route, establishing climb and descent profiles, and adhering to the plan unless a significant cause merited departure. This assured consistency while ignoring real-time opportunities. In contrast, adaptive FMS maintain a constant conversation with external conditions, tracking surrounding aircraft, checking meteorological inputs, and assessing performance depending on load and engine health. Instead of reacting to a delay, they plan, anticipating turbulence, recommending changed descent rates to avoid stacking, or pointing out speedier approach approaches that remain on time.

One documented example of these impacts is the adaptive model presented in the International Journal of Aviation, Aeronautics, and Aerospace, which demonstrated improvements in both fuel efficiency and route flexibility[5]. The findings highlighted that adaptive algorithms were not limited to simulations but also responded effectively under emergency conditions, offering resilience as well as efficiency. These results underline that the impact of adaptive systems is not theoretical—it is measurable in real-world operations and provides a concrete pathway for the next generation of flight management systems. The documented results in visible in Figure 1. and Figure 2.

Figure 1. AI FMS algorithm. The pink blocks is a string constant/input in LabVIEW. It provides textual data ("abc" in this case) that can be passed into other functions or controls within the algorithm. Its contribution to the overall system is supplying identifiers or textual commands that help the algorithm interpret paths, labels, or module names. The "error out" blocks are error cluster indicators. It captures error information from upstream functions (status, code, and source). By wiring this into the main algorithm, it ensures that if any step fails, the error is flagged and propagated, helping maintain robustness and debugging capability. The "Prog" blocks represent a Boolean control (with values True/False, shown as "OK/TF"). It allows the program to check or toggle a condition during execution. Within the overall algorithm, it acts as a trigger to proceed with execution only when the initial system state is valid. Regarding the blocks with the numeric constant 3.8, such constants are often used for calibration, scaling factors, or threshold comparisons. Its contribution is ensuring that this fixed reference value is available wherever needed in the larger control flow. Additionally, the module path block is a file path constant/input. It defines a pathway to a file or module needed by the algorithm. In the context of the bigger system, it ensures that the algorithm knows where to retrieve or save module-related data, making it essential for file management. Finally, the function with "abc" strings and "?!" terminals is a string processing function block that handles multiple text inputs and outputs. It takes several string inputs, processes or formats them, and outputs results for further use. Its role in the overall algorithm is text handling—ensuring data like aircraft identifiers, callsigns, or module labels are standardised for downstream functions[5].

Figure 2. Breakdown flowchart of the algorithm found in Figure 1.

Figure 1 and Figure 2 show the algorithm used between the FMS panel and the “flight info” panel. The session opens when the enter button on the primary FMS panel is clicked and runs all the data from the text input through the Python code. Those codes include API keys from weather monitoring and ChatGPT servers. This information is processed there, and the best possible results are relayed to the flight info panel. This, session is then terminated, and no further information is processed. During the process, however, some other notable features are also present. The best of which is the fact that, based on real-time monitoring data from the weather API key, the emergency section on the flight info panel can show “true” if there is bad weather, for example, and suggest the pilot take another route.

In addition to the algorithms flowchart, the only points of conflict is after the analysis of patterns. This is the point when the pilot must take over to input specific data, but this circumstance is very minimal, as the study concluded the accuracy to be >90%.

In Figure 3. The main FMS panel is found in the middle and displays the flight number, locations, and fuel requirements. This is vital information, or at least the basic information needed to calculate a basic flight plan for the aircraft to follow. The information is taken through the algorithm's input program and uses AI to formulate the output on the left panel that displays flight info. Another feature that was important to mention is that the map on the right can change its route based on weather data from a weather API key found in the Python code.

Figure 3. Flight testing the deck of the algorithm [5].

Discussion

The algorithm described above had accuracy levels >90%, which, for a testing module, was very successful in its first run. During the testing phase, the algorithms worked flawlessly with very minimal problems in the Python scripts, which were quickly fixed. From this information, we can deduce that the algorithm was fit for its purpose and functions as intended. Nevertheless, a significant drawback to this research was that it was done with no funding. Therefore, the development and extent of this research are limited. For example, it could not be tested in the real FMS ecosystem that would really prove this algorithm's capabilities in the cockpit. This is a major aspect of the research that needs to be worked on in the future to test it in a real avionic network.

One suggestion to further improve the algorithm would be to search for and apply for funding or stipends. This will allow us to run real-world simulations inside a real FMS ecosystem and optimise this technology. The algorithm was made on a community license on LabVIEW, which significantly limited its operation. Furthermore, the structure of the algorithm should have been planned and created prior to the investigation, as otherwise it could have turned out to be relatively messy. This could make the algorithm more user-friendly, too, to ameliorate its implementation.

While recognising the limitations of the study, as we saw above, implementing the FMS can have significant effects on the industry. A one or two per cent reduction in fuel consumption may appear insignificant, but when multiplied by hundreds of flights each week, this represents a significant financial change. Airlines that can cut even that much from their operations will save millions of dollars each year while also lowering emissions and satisfying environmental goals. In reality, the International Air Transport Association (IATA) revealed that in 2019, fuel accounted for over 28.7% of worldwide airline operating expenditures, totalling more than $188 billion in yearly spending[6]. This demonstrates why even tiny efficiency gains are highly valued in the aviation business, and by extension, why FMS represent the future of the aviation industry.

Furthermore, with regulators tightening pollution restrictions, identifying efficiency gains that do not disrupt schedules is a competitive advantage. Adjusting the route or altitude in mid-flight minimises CO₂ emissions without passengers noticing. Older fixed-route systems, on the other hand, only allowed for broad adjustments: crews relied on limited forecasting models, resulting in many missed opportunities for improvement[2]. Adaptive flight management systems (FMS) reflect a structural shift, with routing constantly modified by incorporating live weather feeds, traffic data, and engine performance parameters.

Other arguments emphasise the broader implications of adaptive management. Even experienced crews benefit from early notifications. If a problem— whether weather-related, mechanical, or traffic-related—can be resolved minutes earlier, the number of solutions increases. Route changes that avoid unstable air or optimise descent rates can make the travel significantly smoother, which is important for passenger comfort and brand loyalty. Older fixed systems frequently created inefficiencies such as "traffic stacking" during descent or excessive turbulence exposure, which adaptive systems may now avoid by assessing conditions in real time[2].

Nevertheless, FMS are currently experiencing significant delays. One source is Eurocontrol's AIRRAC (Airline Route Availability Circular), which refreshes available routes in controlled airspace every 28 days[10]. This relatively sluggish update frequency forces airlines to operate with data that may be out of date, restricting efficiency and lowering flexibility in everyday operations. However, AIRRAC data remains important for routing possibilities since it offers a baseline legal structure for which paths are acceptable. Integrating adaptive and AI-driven solutions could greatly reduce this timeframe by automating continual changes and dynamically combining real-time weather and air traffic data.

In addition, unlike traditional avionics that require full software upgrades to incorporate improvements, adaptive systems refine their own performance with each flight, learning from operational outcomes and becoming progressively more efficient. This ability to “learn in the loop” is why adaptive FMS are considered the most promising pathway to balancing efficiency and safety in modern aviation. We can clearly see this from the study done in 1989[11]. Although this was a much earlier time, it still demonstrates that the technology adaptive systems (which in this case was the flight control system), were already in development years ago. The study made conclusions on the practicality of these adaptive systems and stated that they made these systems for greater efficiency[11].

Conclusion

An adaptive machine learning system would prove to be the future of aviation[12]. However, actually integrating this into the current FMS environment will take a lot of funding and research before it even reaches the testing phase[13]. Factors like safety (FAA rules), reliability, efficiency, and resource count need to be taken into account before starting a massive project like this[14][15].

Since the FMS is one of the most important components on the aircraft, smaller avionics components can be targeted first. In this way, it will be much more cost-effective in terms of research funding and be more manageable as a smaller component will be easier to deal with and can be a benchmark for further development[16].

Overall, AI is inevitably going to be present on future aircraft [12]. However, each one is going to have its own algorithm and connective branches to ensure that no component works independently of the others[15]. This may take years to finalise and be put into practice, but it will improve the economic and safety sector of future air travel[13][14].

Bibliography

All figures are produced by the author.

[1] ICAO. (2023). Artificial Intelligence in Civil Aviation. https://www.icao.int/safety/AI-in-aviation

[2] National Academies of Sciences, Engineering, and Medicine. (2003). The Status of Aviation Weather Forecasting.https://nap.nationalacademies.org/catalog/10637/the-status-of-aviation-weather-forecasting

[3] Honeywell Aerospace. (2021). Next-Generation Flight Management Systems.https://aerospace.honeywell.com/us/en/learn/products/fms

[4] Collins Aerospace. (2020). Avionics Solutions. https://www.collinsaerospace.com/what-we-do/Industries/Commercial-Aviation/Avionics

[5] Chittayil, S. (2024). Adaptive AI in Flight Management Systems. International Journal of Aviation, Aeronautics, and Aerospace.

[6] IATA. (2019). Fuel Fact Sheet. https://www.iata.org/en/pressroom/fact-sheets/fuel/

[7] Aerospace America (AIAA). (2023). AI: How it’s delivering sharper route 

planning.https://aerospaceamerica.aiaa.org/features/ai-how-its-delivering-sharper-route-planning/

 [8] MIT News. (2025). Learning how to predict rare kinds of failures in automated 

systems.https://news.mit.edu/2025/learning-how-predict-rare-kinds-failures-0521

 [9] Ramée, J. (2020). Evaluation of legacy flight planning modules. Georgia Institute of Technology.

[10] Eurocontrol. (2021). Airline Route Availability Document (AIRAC) / AIRRAC Information. Retrieved from: https://www.eurocontrol.int/service/airac

[11] R. Smith and J. L. Speyer, "An adaptive flight control system based on an explicit parameter identification technique," Proceedings. ICCON IEEE International Conference on Control and Applications, Jerusalem, Israel, 1989, pp. 483-487, doi: 10.1109/ICCON.1989.770564.

[12] Koul, S., Jain, A., & Sharma, P. (2025). A review of machine learning applications in aviation engineering. ResearchGate. https://www.researchgate.net/publication/388813502_A_review_of_machine_learning_applications_in_aviation_engineering

[13] Tafur, D., Lema, J., & Valencia, C. (2025). Applications of artificial intelligence in air operations. Journal of Air Transport Management. https://www.sciencedirect.com/science/article/pii/S2590123024019856

[14] Federal Aviation Administration (FAA). (2024). Roadmap for Artificial Intelligence Safety Assurance. FAA. https://www.faa.gov/aircraft/air_cert/step/roadmap_for_AI_safety_assurance

[15] Azyus, R., Alwan, M., & Singh, K. (2025). Regulatory, ethical, and security dimensions of AI in aircraft maintenance: A framework for assessing harm. ResearchGate. https://www.researchgate.net/publication/390269515_Regulatory_Ethical_and_Security_Dimensions_of_AI_in_Aircraft_Maintenance_A_Framework_for_Assessing_Harm

[16] Berri, P., Mainini, L., & Smith, A. (2020). Computational framework for real-time diagnostics and prognostics of aircraft actuation systems. arXiv Preprint. https://arxiv.org/abs/2009.14645