Interpreting Trainee Intent in Real Time in a Simulation-based Training System

摘要

This article describes a novel methodology for interpreting the actions of trainees in a simulation-based training system. Called Template-based Interpretation (TBI), it dynamically monitors the low-level actions of a trainee in the simulation environment in order to identify his intended high-level action in real time. This is important in simulation-based trainers where an instructor is required to provide feedback to the student. The use of this technique can assist the instructor by lifting the burden of having to continuously monitor and evaluate the trainee's actions. Fundamentally, TBI uses a form of Case-based Reasoning (CBR), matching a series of low-level actions performed by the student against pre-defined templates that describe the typical actions that a human would execute when carrying out a particular intent. However, it differs from the traditional CBR systems in several ways. The most significant is that instead of having a benchmark problem to which the similarity of the several historical cases is measured, TBI simply maintains a set of "open" templates over an indefinite period of time. Each of these open templates becomes either increasingly or decreasingly representative of the student's intents, as judged by how the student's low-level actions "complete" each open template over a period of time. This process is designed to operate in real time, so that feedback can be provided to the student during the training session. The technique was incorporated in a prototype system called TRAMS. TRAMS was applied to a flight-training simulator to identify the pilot's intended actions. The results showed that TBI was not only effective in its intended mission, but was able to do so in a timely fashion. This article also covers the testing program and its results.