Robot-assisted cognitive assessment with eye-tracking

Most robot-assisted cognitive and mental health assessments have relied on self-report questionnaires or standardized cognitive tests. Although these tools yield useful summary scores, they may miss subtle, continuous changes in attention, visuomotor control, and memory. Eye-tracking tasks, such as smooth pursuit and visuospatial memory (VisMET), can provide continuous measures of cognitive and oculomotor processes. Previous studies have compared robot- and human-assisted conventional cognitive assessments, but matched comparisons of robot- and human-assisted eye-tracking workflows remain scarce. Here, we directly compared robot-assisted and human-assisted eye-tracking assessment workflows using the same eye-tracker model and matched procedures. Administrators completed both conditions in a within-subjects design, whereas recipients were assigned to one condition and completed the smooth pursuit and VisMET tasks. Dependent measures included velocity gain and root mean square error for smooth pursuit, and valid gaze sample, fixation, saccade, and area-of-interest measures for VisMET. Recipients and administrators also rated usability, perceived workload, and trust using the SUS, NASA-TLX, and SSRIT. We did not detect statistically significant between-condition differences in eye-movement or user-experience measures among assessment recipients (all FDR-corrected ps ≥ 0.128). Administrators reported lower NASA-TLX scores in the robot-assisted condition than in the human-assisted condition (25.67 vs. 37.16; FDR-corrected p = 0.081, Cohen's dz = −0.50), and usability and trust ratings were similar across conditions. These findings provide preliminary evidence for the technical and workflow feasibility of robot-assisted eye-tracking assessment and support further evaluation in broader, clinically relevant populations.