Towards Application of Adaptive Instructional Systems in Simulation-Based Lifeboat Training Using Bayesian Networks.

Lifeboat coxswains need to be trained and assessed regularly on their performance in plausible emergencies. The ability to do this frequently is limited by the safety risks and logistics of practicing in harsh offshore conditions. Simulation-based lifeboat training provides a means for coxswains to practice typical offshore emergency scenarios, which are usually performed under the supervision of an instructor with occasional interventions. This research serves as a pilot study for automating the role of a human instructor and providing a customized training experience for learners by developing an adaptive instructional system (AIS) for simulation-based lifeboat training. To inform the learner model of the AIS, probabilistic models of learners’ behaviors were developed using Bayesian networks for launching, navigation, and slow-speed maneuvering tasks of a typical lifeboat training exercise. The model is able to evaluate learners’ actions and behaviors to diagnose their skill levels, strengths, and weaknesses, based on the evidence collected while the learners perform a variety of tasks during a training scenario. A case study is presented to demonstrate how the learner model can be trained with simulation-based assessment data and applied to inform a pedagogical model of an AIS to tailor instructional pace, training scenarios, and feedback to the learners’ needs. The results of this study provide an important step towards applying AISs in simulation-based lifeboat training and other maritime safety simulation-based training environments, which are expected to improve learners’ skill acquisition and speed their time to competence.