Detecting Expertise in Gaze Data Using Hidden-Markov-Models

The circular factory concept offers a promising solution to rising global resource demand. To realize this, the Collaborative Research Centre (CRC) 1574 "Circular Factory" is developing a lab-scale setup where the disassembly and reassembly of angle grinders are central tasks. Automating this process is complex, because of the nature of used products, which vary in generation and condition. Therefore, human expertise currently remains essential. To successfully transfer this intrinsic knowledge to robots via learning-from-demonstration, training data must be weighted according to the demonstrator's expertise. This work proposes an automated approach to evaluate human expertise using eye-tracking. We conducted an experiment (N = 38) involving manual disassembly and reassembly of angle grinders, collecting gaze data via a Tobii Pro Glasses 3 system. Gaze points were mapped to the five distinct Areas-of-Interest (AOIs) ‘Instructions’, ‘Bins’, ‘Tools’, ‘Angle Grinder’, and ‘Other’ to generate discrete gaze sequences. Hidden Markov Models (HMMs) are applied to these sequences to classify participants' expertise levels. Currently, data analysis and HMM training are underway. We hypothesize that HMMs can effectively distinguish expertise levels based on gaze patterns, enabling automated selection of high-quality demonstration data. This contributes to bridging the gap between human cognitive skills and highly adaptable robotic automation.