Perceptual versus memory-guided affordances: training-related adaptations in a spatial computer game

Traditional sequence learning research relies on discrete button-press paradigms where pauses segment responses. We developed a continuous game-based adaptation of the Serial Reaction Time task and tracked neural changes in 25 participants using fMRI before and after 1–3 weeks of training. Participants learned to control an avatar’s position and follow fixed and random sequences in the game. Behaviourally, fixed sequences showed robust learning with shorter response times and increased number of perfect playthroughs between Scan1 and Scan2. Random sequence performance unexpectedly declined, with fewer perfect playthroughs, consistent with negative transfer from consolidated sequence representations. For fixed sequences at Scan2 relative to Scan1, we identified the expected cerebellar and occipital decreases alongside unexpected increases in bilateral superior temporal regions. Random sequences showed increased insular activity, consistent with conflict monitoring when consolidated predictions fail. The temporal recruitment suggests continuous sequences consolidated through integrative mechanisms, potentially grammatical encoding, verbal mediation, or spatiotemporal binding, not typically reported in discrete tasks. Critically, training produced no premotor or superior parietal decrease within the fixed sequence, the automatisation signature discrete paradigms report. Our results suggest that discrete sequence learning principles may not fully generalise to continuous environments and highlight the need for paradigms that capture the spatiotemporal demands of real-world skill acquisition.