How the Brain Keeps Conflicting Motor Rules Apart

Imagine learning to ride two bicycles at once: one normal, one reversed. On the normal bike, turning the handlebars left turns the wheel left. On the reversed bike, the same action turns the wheel right. To steer both bikes in the same direction, your brain must hold two completely opposite rules in memory at the same time and apply the correct one in an instant. Humans can do this, and behavioral studies have shown that it is the preparation of movement, rather than the physical movement itself, that keeps these conflicting rules distinct. But where in the brain are these rules stored, and how are they represented? A new study using functional magnetic resonance imaging offers a surprising answer.

Researchers Antoine Caraballo and Sungshin Kim of Hanyang University in Seoul designed an experiment to separate two fundamental processes of motor learning: motor planning, the mental computation of what movement is needed, and motor execution, the physical carrying out of that movement. Thirty-three right-handed adults learned to reach for visual targets on a screen while their cursor feedback was rotated by plus or minus thirty degrees, forcing them to tilt their hand in the opposite direction to compensate. Crucially, the experiment was structured so that participants sometimes planned different rotations but executed the exact same physical movement, allowing the researchers to isolate the neural signature of planning from that of execution.

The task unfolded across five fMRI runs. In the first two runs, participants adapted to three sensorimotor mappings indicated by distinct contextual color cues: a baseline mapping with no rotation, a counterclockwise rotation of minus thirty degrees, and a clockwise rotation of plus thirty degrees. All targets appeared in the same central position, so the only thing distinguishing the conditions was the learned rule linking the color cue to the required hand movement. By the end of the second run, participants’ movement directions matched their goal angles almost perfectly, confirming that they had successfully learned all three conflicting mappings simultaneously.

The next three runs were designed to dissociate the components of visuomotor adaptation. In Run 3, the target position stayed fixed while the applied rotations varied, so both the planned rotation and the physical movement direction changed together. In Run 4, the researchers pulled off a clever trick: target locations and rotations were offset so that they cancelled each other out, meaning participants planned different rotations for different visual targets but executed the exact same horizontal hand movement every time. In Run 5, all rotations were removed, so participants simply reached directly for targets appearing at different positions. By comparing brain activity across these runs, the researchers could isolate three components: the motor plan generated to counteract a rotation, the physical execution of the movement, and the processing of the visual target location.

The univariate results showed increased BOLD activity in the posterior parietal cortex, or PPC, during rotational planning, supporting the region’s involvement in processing learned sensorimotor mappings. This aligns with prior studies showing increased PPC activity during the late stages of visuomotor adaptation, consistent with a role for the PPC in representing, retrieving, applying, or storing learned mappings. Interestingly, no significant univariate activity was observed in the cerebellum, which the researchers attribute to the participants’ advanced stage of learning: by the time the critical runs took place, participants had already adapted, so the number of sensory prediction errors was minimal.

But the more provocative findings came from multivoxel pattern analysis, or MVPA, a technique that decodes information from the spatial pattern of activity across many voxels rather than from average signal strength. Using a cross-classification approach, the researchers trained a classifier on one set of conditions and tested it on a different set, ensuring that any decoded information reflected an invariant representation rather than a context-specific confound. The results revealed a clear dissociation between motor execution and rotational planning.

For motor execution, the PPC and cerebellum showed distinct decodable neural patterns for opposing movements. Significant above-chance decoding accuracy was found in the contralateral primary motor cortex, primary somatosensory cortex, left inferior parietal cortex, visual area, and right cerebellar lobule VI. This means that even when the visual target location remained identical, these regions encoded the direction of the physical movement at a fine-grained level. Conversely, for rotational planning, no region showed significant decoding of the opposing plans. The cerebellar ROIs failed to discriminate between the neural plans made to counteract the clockwise and counterclockwise perturbations, and no other predefined ROI showed above-chance decoding either.

This null result is particularly surprising given the PPC’s established role in storing updated sensorimotor mappings, and it contrasts with previous studies that successfully decoded rotational planning in sensorimotor regions. The researchers suggest several explanations. Earlier designs did not fully counterbalance all three visuomotor components, so their decoding results may have partially reflected differences in physical hand movement direction, a confound the current design minimized. Alternatively, the relatively larger rotations in previous studies, ninety degrees and forty degrees compared to thirty degrees here, likely engaged more explicit cognitive strategies, resulting in more distinct activation patterns. The neural representation of rotational planning may also be sensitive to the tool used: earlier studies used a joystick involving wrist movements, whereas this design employed a tablet requiring whole-arm reaching.

The authors acknowledge three notable limitations. First, only three functional runs were included for the dissociation analysis, resulting in a relatively small sample of roughly thirty to thirty-six trials per condition, which fundamentally constrains the statistical power of the multivariate decoding. Because rotational planning involves subtle neural activity, this sample size may lack the sensitivity required to detect such spatial representations. Consequently, the non-significant decoding should be interpreted as a lack of robust decoding evidence rather than definitive evidence for the absence of rotational planning representations. Second, participants initiated their reaching movements as soon as the target appeared, without a preparatory interval, allowing the retrieval of the adaptation plan and motor execution to overlap in time. Future studies introducing an explicit preparation time using a Go/No-Go paradigm would separate planning from execution at the behavioral level. Third, although target color cues were counterbalanced across participants, the cue-condition mappings remained present across runs, so the visual area decoding may partly reflect low-level visual features such as color.

In summary, this study presents new insights into the neural substrates underlying visuomotor adaptation. The primary visual cortex can distinguish spatial target locations, reflecting its well-established retinotopic organization. After participants learned to counteract the rotations, neural activity within the parietal cortex increased, implying that this region is central to processing learned sensorimotor mappings. However, while the cerebellum and PPC demonstrated decodable neural patterns for opposing motor execution, no significant decoding was observed for opposing rotational planning. This raises open questions about how distinct parameters of visuomotor mappings are represented in the human brain, and whether the PPC stores learned rules in a format that is not yet accessible to current decoding techniques, or whether the representations are simply too subtle to detect with the trial counts used here.

Subject of Research: Dissociating neural representations of multiple learned sensorimotor mappings

Article Title: Dissociating neural representations of multiple learned sensorimotor mappings

Article References: Caraballo, A., & Kim, S. (2026). Dissociating neural representations of multiple learned sensorimotor mappings. Frontiers in Behavioral Neuroscience, 20, Article 1965610. https://doi.org/10.3389/fnbeh.2026.1965610

Image Credits: AI Generated

DOI: 10.3389/fnbeh.2026.1965610

Keywords: Dissociating, neural, representations, multiple, learned, sensorimotor, mappings, scientific research