From experience to action: sequential dynamics of experiential learning in immersive virtual reality-supported safety training

We examine the temporal dynamics of hazard recognition learning in Immersive Virtual Reality (IVR), addressing limitations of outcome-based research that treats cognitive mechanisms as a “black box”. Grounded in Experiential Learning Theory (ELT), we employ a microgenetic framework utilising high-density behavioural logging and concurrent think-aloud protocols from 27 civil engineering students. Using Hidden Markov Models (HMM) and sequential analysis, we identify 12 behavioural transitions that survived multiple-comparison correction (p adj < .05) that correspond to ELT stages. An exploratory typology of three learning trajectories (Successful Progression, Iterative Refinement, and Sustained Disruption) emerged, mapping specific behavioural pathways, including the translation of scaffolding into action and formation of problematic disruption loops. The findings motivate three theoretically grounded extensions of ELT for immersive contexts: (a) cycle arrest, in which learners stall within a single mode; (b) probabilistic stage transitions, in which movement between modes is stochastic rather than deterministic; and (c) trap states, self-reinforcing loops that resist recovery without external intervention. We provide a methodological framework for process-oriented IVR analysis, offering granular behavioural evidence to inform AI-driven adaptive scaffolding, trajectory-aware feedback, adaptive intervention timing, and evidence-based instructional design.