Current trends in simulation-based military training show an increasing demand for artificial intelligence (AI) and data science technologies to offer more flexible and adaptive training solutions for military personnel in rich simulated training environments. This requires technologies capable of (1) measuring and assessing performance of trainees in real-time, and (2) simulating role-playing agents that can replace human role-players, and adapt their behavior to guide learning experiences for a trainee.

The above requirements pose two challenges. First, current training systems often lack an understanding of the dynamic context of a trainee’s behavior, which is required to judge its performance during training of specific missions, (part-)tasks or tactical situations. This context is typically only available in the head of an observing instructor. Second, behavior models for simulated role-players are often black boxes from the point of view of an instructional system and cannot easily be adapted for instructional purposes (e.g. exhibiting degraded performance or making deliberate mistakes).

In this paper we present a unified human behavior modelling (HBM) approach that addresses the above challenges. It is based on the idea that HBMs can be used to model roles for both human trainees and agents. For an agent, it acts as an AI model that produces behavior, equipped with predetermined adaptive variables. For a trainee, it acts as an observer that tracks, measures and assesses behavior being performed by a trainee. The HBMs (1) support instructor-based and automated tutoring, (2) promote collaborative design of instructional systems between training designers, subject-matter experts and behavior modellers, and (3) allow for interchangeable roles in training.

We demonstrate a proof-of-concept of the HBM approach in the scope of a training system for military aircrew training. Several use cases for adaptive training for fighter pilots are addressed and based on an implementation, challenges and lessons learned are discussed.