The Eyes Don’t Lie: Detecting Cognitive Load in VR Assembly Training using Built-in Pupillometry

Abstract

Careful design of Virtual reality (VR) training is necessary to optimize knowledge retention while avoiding excessive mental strain. In this context, personalized learning trajectories represent a key strategy by dynamically addressing individual needs, such as adapting the pace, content or level of guidance. In such trajectories, cognitive load metrics provide a reliable indicator of perceived task difficulty, serving as a trigger to adjust the training parameters accordingly. However, current cognitive load assessment methods are often intrusive or limited to post-hoc subjective measurements. Therefore, we assessed whether pupillary responses captured by built-in eye-tracking sensors reflect cognitive load. Using a comparative experimental design, we manipulated task difficulty through varying assistance levels in a VR assembly training to compare pupillary responses against subjective workload differences confirmed by the NASA-TLX. The results of our study demonstrate a clear convergence between subjective workload and objective physiological data: increased perceived mental effort was mirrored by significant pupil dilation, while a decrease led to pupil constriction. These findings support the feasibility of built-in pupillometry for cognitive workload detection and provide a necessary step toward adaptive VR training systems that respond to the trainee’s cognitive state in real time.