Continuous Game Training Reveals Unexpected Brain Mechanisms Underlying Motor Sequence Learning

A longitudinal neuroimaging investigation has revealed that mastering continuous motor sequences engages distinct neural circuits and produces behavioral interference patterns that diverge from traditional discrete learning paradigms. Reporting in Frontiers in Neuroscience, researchers Olgierd Borowiecki and Alicja M. Olszewska examined how the human brain acquires and consolidates sequential actions when movements flow continuously through space and time without artificial trial segmentations. Using a newly developed avatar-based computer game coupled with functional magnetic resonance imaging across two scanning sessions separated by home training, the authors evaluated twenty-five healthy adult volunteers to track how sensory-driven reactions transform into memory-guided motor trajectories. Their primary neuroimaging analysis showed that after extended sequence training, execution of a predictable sequence did not decrease activation across dorsal premotor and superior parietal regions, but instead elicited unexpected bilateral activity increases within superior temporal cortices alongside selective cerebellar disengagement. Concurrently, behavioral measurements demonstrated robust consolidation for repeating sequences accompanied by an unexpected performance collapse during unpredictable sequences, uncovering a form of negative transfer rarely documented in standard laboratory skill assessments.

For decades, cognitive neuroscientists studying procedural memory have relied on classical Serial Reaction Time tasks, where individuals respond to discrete visual cues appearing at predetermined locations on a computer monitor. In these traditional experiments, participants press dedicated keys corresponding to each stimulus, and successive actions remain separated by brief pauses or visual gaps that segment individual trials. While these discrete paradigms established fundamental principles regarding how the human nervous system extracts statistical regularities, they hold action alternatives simultaneously visible and separate each movement into an isolated decision. Real-world motor skills such as driving along an obstacle-strewn roadway or navigating dynamic environments require uninterrupted coordination, where early maneuvers continuously shape future movement possibilities. To bridge this ecological gap, the researchers designed an experimental environment where perceptual affordances—immediate sensory opportunities for action—could progressively transform through repeated practice into memory-guided affordances that anticipate environmental geometry before sensory confirmation becomes available.

The experimental task was engineered in the Unity engine as a continuous, vertically scrolling two-dimensional maze viewed from an overhead perspective, reminiscent of mechanics found in retro video games. Participants operated MRI-compatible response grips using their left and right thumbs to steer a blue square avatar along the horizontal axis among three distinct lanes while the environment scrolled downward at a constant velocity. Success required steering into open corridors to collect high-value green rewards while avoiding red penalty items across sixteen consecutive corridor junctions within each nineteen-second playthrough. The optimal navigation path yielded a maximum score of two hundred and sixty points. At each decision point, participants faced a brief visibility window of approximately 0.8 seconds to evaluate oncoming corridors and adjust avatar positions. To isolate structure-specific learning from baseline sensorimotor familiarity, the experimental protocol interleaved fixed sequences, which featured an unvarying spatial sequence of target corridors across every repetition, with random sequences, whose corridor patterns and reward distributions were generated uniquely on every playthrough.

To evaluate skill progression, forty-seven young, right-handed individuals aged nineteen to thirty-two were initially recruited, with prior video gaming experience required during initial screening. Following familiarization, participants completed an initial neuroimaging session, designated Scan1, before undergoing intensive home practice. Progression to the post-training scanning session, Scan2, required achieving at least two-thirds perfect playthroughs on the fixed sequence during the final home training session within an allotted training window of up to three weeks. Thirty-one participants completed Scan2, and data from twenty-five participants aged nineteen to twenty-nine years, comprising thirteen females and twelve males, were retained for final analysis after excluding individuals impacted by an early software configuration issue. Twenty participants achieved the training criterion within one week of practice, four participants required two weeks, and one volunteer required three weeks. The home training software automatically recorded performance logs, verifying that participants completed one hundred and fifty sequences per week, evenly divided between fixed and random conditions, under strict daily submission deadlines.

Behavioral analyses demonstrated clear learning of the fixed sequence accompanied by pronounced negative transfer on random sequences. At Scan1, participants averaged 10.00 perfect playthroughs with a mean response time of 703.16 milliseconds on fixed sequences, compared to 8.24 perfect playthroughs and a mean response time of 721.52 milliseconds on random sequences. Following home training, fixed sequence performance at Scan2 climbed to an average of 13.84 perfect playthroughs while response times dropped precipitously to 260.94 milliseconds, representing an anticipatory execution state that fell well below the 0.8-second perceptual threshold required for visual confirmation. In striking contrast, performance on random sequences deteriorated significantly from Scan1 to Scan2, dropping from 8.24 to 3.88 perfect playthroughs, while response times decreased only slightly to 668.59 milliseconds. A repeated-measures analysis of variance confirmed a significant scan-by-condition interaction for both perfect playthroughs and response times, confirming that extended training on fixed trajectories impaired participants’ capacity to process unpredictable environmental layouts.

Neuroimaging data were acquired on a 3 Tesla scanner using an interleaved gradient-echo echo-planar imaging sequence sensitive to blood-oxygen-level-dependent contrast, complemented by high-resolution structural scans. Functional data underwent standardized preprocessing through fMRIPrep and ICA-AROMA motion artifact denoising, followed by two-level general linear model evaluations in SPM12 corrected for multiple comparisons at a cluster-level family-wise error threshold of p < 0.05. Whole-brain contrasts during Scan1 captured an exploration state: fixed sequences elicited greater bilateral frontoparietal activation than random sequences, engaging the left inferior parietal lobule, right superior and middle frontal gyri, and right precuneus. This frontoparietal cognitive control network reflected explicit, effortful pattern search and working memory deployment during initial game exposure. Within Scan1, direct comparisons showed that while middle frontal activations were prominent during the first session, they attenuated by the third session, leaving primarily parietal clusters engaged as participants adapted to the immediate task demands.

Following home practice, whole-brain neuroimaging contrasts between Scan1 and Scan2 isolated the neural reorganisation underlying consolidated continuous navigation. In the fixed condition, prolonged training produced pronounced activation reductions within the bilateral inferior, middle, and lingual occipital cortices, calcarine sulcus, cerebellar lobule VI, cerebellar lobule VIII, vermis lobules VI and VII, and right insula. Concurrently, the fixed sequence elicited significant longitudinal activation increases within bilateral superior temporal gyri and the precuneus extending into the posterior cingulate cortex. Contrary to prior findings from discrete button-press paradigms, no significant decreases emerged in dorsal premotor cortex or superior parietal lobule for fixed sequence execution. Furthermore, fixed-versus-random sequence contrasts revealed no significant striatal clusters at either scanning session, and the anterior cingulate cortex exhibited no significant task-related modulation across conditions, showing that classical discrete automatisation models do not fully capture continuous task execution.

The neural profile observed during random sequence performance at Scan2 provided evidence regarding the cognitive cost of violated expectations. Contrasting random sequences against fixed sequences at the second scan identified a widespread activation network encompassing the right insula, left superior frontal gyrus, vermis lobules VI, VII, VIII, and IX, bilateral cerebellar lobule VI, bilateral supplementary motor area, and left middle and inferior occipital regions. Furthermore, longitudinal tracking revealed that the right insula showed increased blood-oxygen-level-dependent activity from Scan1 to Scan2 specifically during random sequences, whereas fixed sequence execution exhibited reduced insular activity over the same period. A significant time-by-condition interaction localized primarily to the right insula and cerebellar cortex. The investigators interpreted elevated insular and supplementary motor area responses during random navigation as evidence of conflict monitoring and internally guided movement generation triggered when a consolidated motor program repeatedly collides with unpredictable spatial geometry.

The authors discussed several non-mutually exclusive theoretical mechanisms to explain the unexpected recruitment of bilateral superior temporal gyri during consolidated fixed sequence navigation. Because continuous navigation requires binding sixteen sequential choices into a coherent spatiotemporal trajectory across extended durations, temporal regions implicated in relational processing, syntax, and grammatical abstraction may consolidate these complex action streams. Alternatively, participants may have leveraged covert verbal labels, such as categorizing movements into sequences of left, middle, or right, to scaffold motor chunking across predictable timeframes. A third mechanism posits unified spatiotemporal binding, integrating directional and motor parameters across uninterrupted movement paths. Finally, the authors noted that the rhythmic 1.16-second periodicity between corridors could have entrained superior temporal circuits sensitive to auditory and sensorimotor timing predictions. Future investigations employing dual-task articulatory suppression, cross-effector transfers such as foot pedals, and disrupted interval timings will be required to distinguish between these candidate mechanisms.

The researchers underscored several methodological boundaries and design considerations that frame the interpretation of their results. The continuous task deliberately combined four co-occurring design elements: uninterrupted movement flow, explicit cueing of sequence repetition, three-lane navigational steering, and bimanual thumb control. Because these factors operated simultaneously to emulate ecological conditions, isolating their individual contributions requires testing targeted variants that selectively alter pacing, awareness, or motor effectors. Additionally, the experimental cohort comprised a modest sample of twenty-five right-handed young adults who had prior gaming background and successfully achieved home training benchmarks within three weeks, meaning the findings reflect high-performing learners under cued-discovery conditions rather than unconstrained populations. While the absence of striatal clusters and premotor decreases must be interpreted cautiously given the sample size, the convergent behavioral negative transfer and distinct superior temporal recruitment demonstrate that continuous environments engage unique neurocognitive pathways that discrete trials cannot observe.

Subject of Research: Neural mechanisms and behavioral dynamics of continuous sequence learning in a spatial computer game

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

Article References: Borowiecki, O., & Olszewska, A. M. (2026). Perceptual versus memory-guided affordances: training-related adaptations in a spatial computer game. Frontiers in Neuroscience, 20, Article 1736859. https://doi.org/10.3389/fnins.2026.1736859

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.1736859

Keywords: neuroscience, sequence learning, functional MRI, motor skill acquisition, cognitive control, serial reaction time, superior temporal gyrus, cerebellum, Perceptual, versus, memory-guided, affordances