Mathematical modeling of growth processes in pilot upgrade training

Pilot training and operational decision-making rely heavily on human experience and situational awareness. Through joint research with airlines, JAXA makes use of training data to model pilot learning, judgment, and behavior by applying statistical and mathematical methods.

This research aims to enhance the efficiency and safety of pilot training and education through a scientific understanding of human learning and development. In the future, this research will serve as fundamental research that contributes to the training and education of not only pilots but also professionals in other fields requiring advanced skills, to the realization of "Human-AI Teaming," where humans and AI work together.

Background and objectives

Even with advances in automation technology, humans still make the final decisions in aircraft operations. Traditionally, training has relied on instructors’ experience and intuition, with little quantitative understanding of questions such as “at what stage do skills develop?” or “which factors influence judgment?” Training is now shifting from time-based technical skill assessments to competency-based evaluations that measure behavioral characteristics, including non-technical abilities like situational awareness. This shift makes it increasingly challenging to assess proficiency. JAXA aims to develop a safe and efficient training system by applying expertise in mathematical statistics and human factors research to better understand human growth and judgment.

To this end, we will conduct research with the following four objectives:
• Quantitatively model skill and judgment growth patterns during training
• Analyze behavioral and cognitive data to quickly identify growth stagnation and characteristics
• Develop a data-based training support and education evaluation system
• Explore future applications in other professions, such as air traffic controllers and astronauts

Achievements to date

JAXA is conducting joint research in collaboration with Japanese airlines to mathematically model the growth process of pilot upgrade training.

Utilizing data accumulated in the Competency-Based Crew Training (CBCT) operated by the airlines, the process by which “improvement required” flags are generated is analyzed as a probabilistic model. The research succeeded in quantitatively capturing the progress of growth from the early to later stages of training (Probabilistic Modeling of Variation in Pilot Performance During Flight Training). This achievement is expected to help training instructors quickly recognize signs of improvement and training effects in trainees, helping them to implement more effective training (see press release "Promoting digital transformation (DX) in pilot training" ).

In the future, in addition to further improving the accuracy of the model, we will continue research and development with the goal of implementing it in actual operations as follows:
• Building a system that updates growth models in real time and connects them to training support AI
• Expanding applications to other fields, such as drone operations, air traffic control, and astronaut training
• Advancing research toward “Human-AI Teaming”, where humans and AI work together to maintain safety.

April 8, 2026