Multiple processes independently predict motor learning

We recruited healthy, young adults (18–35 years old) from the University of South Carolina and surrounding areas. Participants were excluded if they reported any history of a central or peripheral neurological disorder or an ongoing musculoskeletal issue affecting either arm or hand. The study protocol was approved by the University of South Carolina’s Institutional Review Board and all participants provided informed consent to participate.

Data were collected with a bilateral, upper-limb robot (KINARM EndPoint Lab, KINARM, Kingston, Canada) and monocular eye-tracker (EyeLink 1000, SR Research Ltd., Ottawa, Canada) that were integrated with an augmented-reality workspace (Fig. 1a) [41]. Participants sat in a custom chair that used floor-mounted tracks and hydraulics to align them with a forehead rest, which stabilized the head for eye tracking. Participants grasped two near-frictionless manipulanda, which allowed them to make two-dimensional hand movements within an 80 cm wide by 80 cm deep workspace. An opaque shield and fabric cover prevented direct vision of the hands and arms. Hand and gaze position in the robotic workspace were respectively sampled at 1000 and 500 Hz, recorded at 200 Hz, and filtered offline using a low-pass filter with a 20 Hz cutoff.

The augmented-reality environment was created in the same horizontal plane as the robotic workspace by using an inverted-monitor to project visual stimuli at 60 Hz through a semi-transparent mirror. Cartesian gaze position in the horizontal plane was estimated using proprietary calibration algorithms (Kinarm, Kingston, Canada) that provided accurate eye tracking within a workspace of approximately 50 cm wide by 50 cm deep. All visual stimuli were presented within this portion of the robotic workspace. A nonlinear mapping corresponded to a visual area approximately 55° wide by 40° deep in which stimuli located closer to participants comprised larger visual angles.

Participants practiced six trials of a continuous, visuomotor task, Object Hit and Avoid (OHA) [42], once a week for six consecutive weeks. Each participant was scheduled at a consistent time of day on the same weekday to avoid potential confounds caused by circadian rhythms and to assure a consistent retention interval between sessions. Illumination of the room was maintained at a constant level for the duration of the study.

In each trial of the OHA task, 300 red objects comprised of eight geometric shapes (e.g., square, circle, triangle, etc.) moved from the back of the workspace towards the participants along ten parallel paths (5 cm center-to-center spacing) (Fig. 1b). Two shapes were predefined as “Targets” and six shapes were predefined as “Distractors”. Each parallel path contained 20 Targets (n = 200) and 10 Distractors (n = 100) that were released in random order. The average number of objects that were simultaneously present in the workspace and the average speed that objects moved progressively increased over time. As a result, task difficulty increased within each trial, whereas the overall difficulty of each trial was consistent. Each trial ended after all 300 objects had passed through the workspace (~ 2 min).

Participants received standardized instructions to use two green paddles (2.5 cm wide) located on top of each hand to hit away as many Targets and to avoid hitting as many Distractors as possible. When participants made paddle contact with Targets, the robot applied a small perturbation (10 Newtons for 50 ms) to the participant’s hand and Targets rebounded from the paddle with the same direction and speed as the paddle movement. When participants made paddle contact with Distractors, no perturbation was applied to the participant’s hand and Distractors passed unaltered through the paddle. Paddle size, object size and the spacing between adjacent paths prevented participants from simultaneously hitting two objects with the same hand.

We employed six distinct variants of targets and distractors to prevent overlearning of a specific variant from causing plateaus in task performance (Fig. 1c). Each variant was pseudo-randomized and counter-balanced between participants each week and was never practiced by a participant in more than one week. Specifically, each of the six variants was assigned to three different participants each week, and each participant performed six trials of a different variant each week. Before starting each trial, the two target shapes were presented in the middle of workspace until participants confirmed that they had memorized the shapes and were ready to begin. After each trial, participants were offered a rest period until they were ready to start the next trial.

Gaze data were processed and classified using the procedures of a validated methodology for processing gaze data our group previously published [41]. In brief, the methodology involves preprocessing gaze data to remove blink artifacts, one sample spikes caused by incorrect corneal detection, and outliers that occurred when gaze moved outside the eye-tracking workspace. We subsequently use a novel geometric method to transform gaze position data into rotational kinematics of the eye. Finally, we use adaptive thresholding methods to classify eye movements into saccades (rapid eye movements between targets) and smooth pursuits (eye movements that followed moving targets with foveal vision). Our previous manuscript demonstrated that our methodology for gaze processing and classification correctly classifies approximately 90% of saccades and smooth pursuits and misclassifies approximately 5% of saccades and smooth pursuits when compared with manual classification (gold standard) [41].

We used hand and gaze data to compute measures of Task Performance, Skilled Limb Movement, Visual Search, Eye-Hand Coordination and Visuomotor Decisions for each OHA trial.

Task performance We computed two measures of task performance (Eqs. 1 and 2). Targets Hit (%) quantified goal achievement resulting from successful execution of hand movements to hit targets (motor execution). It was calculated as the percent of all 200 targets that participants “hit”, where a target was counted as “hit” if either paddle made contact with the target, causing it to move toward the back of the workspace. Only one “hit” was counted if a target was hit more than once. Distractors Avoided (%) quantified goal achievement resulting from successful inhibition of hand movements to avoid distractors. It was calculated as the percent of all 100 Distractors that were “not hit”, where a distracter was counted as “not hit” if neither paddle made contact with the distractor or if a paddle made contact but caused the distractor to move toward the front of the workspace.

Skilled limb movement We computed five measures of skilled limb movement (Eqs. 3–7). Mean Hand-Speed (cm/s) quantified the overall execution speed of all hand movements by computing the average speed of right- and left-hand movements. Mean Hand-Area (cm²) quantified the overall spatial distribution of all hand movements by calculating the average area covered by right- and left-hand movements, where each area was obtained by computing the convex hull of left- and right-hand movements. Target Contact Speed (cm/s) quantified the execution speed of skilled hand movements by computing the average speed of hand movements at the onset of paddle-contact with each target that was successfully hit. Hand-Speed Bias quantified bimanual coordination by computing inter-limb differences in movement speed. It was calculated as the normalized difference between the average speed of right- and left-hand movements. Hand-Area Bias quantified bimanual coordination by computing inter-limb differences in the spatial distributions of hand movements. It was calculated as the normalized difference between the area covered by movements of the right and left hands. Values of hand-speed bias or hand-area bias near zero indicate equal use of both hands and increasingly higher values indicate greater use of one hand than the other. We were unable to quantify many traditional measures of skilled limb movement, such as time to peak velocity, peak acceleration or smoothness, because we could not identify a distinct start or end point of most limb movements due to the continuous nature of our task.

Visual search We computed three measures of visual search (Eqs. 8–10). Objects Foveated (%) quantified the overall efficiency of visual search by calculating the percent of all 300 objects that participants “foveated” with pursuit eye movements, where an object was counted as “foveated” if the object was followed with foveal vision for at least 40 ms [41]. If an object was foveated more than once, it was only counted once. Spatial Foveation Bias quantified spatial biases in the distribution of visual search by computing the normalized difference between the number of objects foveated on the right and left sides of the workspace. Extrafoveal Hits (%) quantified covert use of parafoveal and peripheral vision for visual search by calculating the percent of targets that were hit but were not previously foveated. We were unable to compute other measures of visual search because a large number of catch-up saccades during pursuit prevented accurate calculation of other valid measures.

Eye–hand coordination We computed two measures of eye-hand coordination (Eqs. 11–12). Gaze-Hand Distance (cm) quantified spatial coupling between the eyes and hands by calculating the distance between gaze and hand position at the onset of paddle-contact with each target [33]. Gaze-Hand Latency (ms) quantified temporal coupling between eyes and hands by calculating the interval between the initial time of each target hit and final time that gaze foveated the target [33‐36, 43]. If a target was hit more than once, only the first hit was included in these calculations. If a target was not foveated or was hit before it was foveated, it was excluded from these calculations.

Visuomotor decisions We computed three measures of visuomotor decisions (Eqs. 13–15). Target Foveation Time (ms) quantified the amount of time used for making decisions to hit targets and was calculated as the average duration that participants foveated targets. Distractor Foveation Time (ms) quantified the amount of time used for making decisions to avoid distractors and was calculated as the average duration that participants foveated distractors. If a target or distractor was foveated more than one time, we included the total time of all foveations. Both measures quantified the average time used to recognize and classify shapes as a target or distractor. However, Target Foveation Time included the average time used to initiate hand movements, whereas Distractor Foveation Time included the average time used to inhibit hand movements. Foveation Time Difference (ms) quantified differences between the amount of time used for making decisions to hit targets and avoid distractors and was calculated as the difference between target and distractor foveation times.

All analyses were performed using Matlab 2017b (Mathworks Inc., Natick, MA).

We enrolled 18 healthy, young adults (8 male, 10 female; 24.2 ± 3.7 years old; 17 R-handed, 1 L-handed) in the study. One participant was unable to complete the sixth week of the study. We included the participant’s data without replacement of the sixth week.

Figure 4 illustrates pursuit and saccadic eye movements (pink and gold lines) and left- and right-hand movements (blue and red lines) made by an exemplar participant at four time points, Week₁·Trial₁ (a), Week₁·Trial₆ (b), Week₆·Trial₁ (c), and Week₆·Trial₆ (d). At each time point, the participant’s eye movements covered an area of approximately 50 cm wide (X) by 40 cm deep (Y). The center-of-mass was consistently located near the midline but shifted distally from around 30 cm on Week₁·Trial₁ (a), to 35 cm on Week₁·Trial₆ (b), and 40 cm on Week₆·Trial₁ and Week₆·Trial₆ (c, d). Combined movements of both hands covered an area that was around 50 cm wide and consistently centered near the midline. However, the range of hand movements in depth increased from around 15 cm on Week₁·Trial₁ (a) to 20 cm on the other three trials (b–d). The center-of-mass also shifted distally from under 10 cm on Week₁·Trial₁ (a) to over 15 cm on the other three trials (b–d). Left- and right-hand movements covered similar areas and were largely constrained to their respective sides.

Figure 4 also displays grids of rectangles that represent each Target (upper grids: 20 × 10) and Distractor (lower grids: 10 × 10) that was foveated and hit (left hand: dark blue, right hand: dark red), foveated but not hit (grey), not foveated but hit (left hand: light blue, right hand: light red), or neither foveated nor hit (white). The participant failed to foveate several targets and distractors on Week₁·Trial₁ (e) but foveated the majority of targets and distractors on the other three trials (f–h). Similarly, the participant failed hit a number of targets on Week₁·Trial₁ (e) but hit the majority of targets on the other three trials (f–h). In contrast, the participant hit several distractors in the first week (e, f) but very few in the last week (f–h). At all four time points, the participant hit more targets with the right-hand, including several targets on the left side of the workspace.

Targets Hit and Distractors Avoided exhibited a low correlation (\(r\)= 0.03), indicating that they quantified unique aspects of task performance. Both measures were included in our subsequent analyses. We also examined each pair of predictor measures for high correlations (\(|r|\)≥ 0.707) indicative of redundant information (Table 1). Two pairs exhibited high correlations, Mean Hand-Speed and Target Contact Speed (\(r\)= 0.89) and Gaze-Hand Distance and Gaze-Hand Latency (\(r\)= 0.89). Target Contact Speed and Gaze-Hand Latency were excluded from all remaining analyses because they had the highest coefficients of variance in each pair [44].

Before testing our two hypotheses, we first confirmed that our participants demonstrated trial-by-trial acquisition and week-by-week retention of improvements in task performance (Table 2). Targets Hit exhibited moderate trial-by-trial increases (\({\beta }_{1}\)=0.26, \({f}^{2}\)=0.23, \(p\)<10^–6), large week-by-week increases (\({\beta }_{2}\)=0.49, \({f}^{2}\)=0.82, \(p\)<10^–6), and small trial-by-trial decreases across weeks (\({\beta }_{3}\)=-0.16, \({f}^{2}\)=0.09, \(p\)<10^–6) (Fig. 5a). We also observed small, week-by-week increases on Distractors Avoided (\({\beta }_{2}\)=0.20, \({f}^{2}\)=0.11, \(p\)<10^–6) (Fig. 5b). These finding show that practice-related improvements in motor execution (Targets Hit) and motor inhibition (Distractors Avoided) contributed to improvements in task performance.

We tested our first hypothesis by examining trial-by-trial acquisition and week-by-week retention of refinements on our measures of skilled limb movement, visual search, eye-hand coordination and visuomotor decisions (Table 2). Three measures of skilled limb movement (Mean Hand-Speed, Hand-Speed Bias, and Hand-Area Bias) displayed practice-related refinements. Mean Hand-Speed exhibited small trial-by-trial increases (\({\beta }_{1}\)=0.09, \({f}^{2}\)=0.03, \(p\)<10^–4) and small week-by-week increases (\({\beta }_{2}\)=0.21, \({f}^{2}\)=0.14, \(p\)<10^–6) (Fig. 6a). Hand-Speed Bias demonstrated small week-by-week increases (\({\beta }_{2}\)=0.14, \({f}^{2}\)=0.03, \(p\)<10^–5) and Hand-Area Bias showed small trial-by-trial increases (\({\beta }_{1}\)=0.14, \({f}^{2}\)=0.03, \(p\)<10^–4). Two measures of visual search (Objects Foveated and Extrafoveal Hits) exhibited practice-related refinements. Objects Foveated displayed small trial-by-trial increases (\({\beta }_{1}\)=0.12, \({f}^{2}\)=0.04, \(p\)<10^–6), moderate week-by-week increases (\({\beta }_{2}\)=0.32, \({f}^{2}\)=0.29, \(p\)<10^–6), and small trial-by-trial decreases across weeks (\({\beta }_{3}\)=− 0.12, \({f}^{2}\)=0.04, \(p\)<10^–6) (Fig. 6b). Extrafoveal Hits exhibited small trial-by-trial increases (\({\beta }_{1}\)=0.22, \({f}^{2}\)=0.12, \(p\)<10^–6) and large week-by-week increases (\({\beta }_{2}\)=0.38, \({f}^{2}\)=0.36, \(p\)<10^–6) (Fig. 6c). Our only measure of eye-hand coordination, Gaze-Hand Distance, demonstrated large week-by-week increases (\({\beta }_{2}\)=0.33, \({f}^{2}\)=0.69, \(p\)<10^–6) (Fig. 6d). All three measures of visuomotor decisions (Target Foveation Time, Distractor Foveation Time and Foveation Time Difference) displayed practice-related refinements. Target Foveation Time showed moderate week-by-week decreases (\({\beta }_{2}\)=− 0.27, \({f}^{2}\)=0.20, \(p\)<10^–6) (Fig. 6e). Distractor Foveation Time displayed small week-by-week increases (\({\beta }_{2}\)=0.11, \({f}^{2}\)=0.04, \(p\)<10^–6). Foveation Time Difference exhibited moderate week-by-week decreases (\({\beta }_{2}\)=− 0.41, \({f}^{2}\)=0.25, \(p\)<10^–6) (Fig. 6f). One measure of skilled limb movement (Mean Hand-Area) and one measure of visual search (Spatial Foveation Bias) did not exhibit practice-related refinements and were excluded from further analyses.

We initially used bivariate regression to identify predictor measures that were individually related to improvements on our two measures of task performance (i.e., at least a small effect size, \({f}^{2}\)≥0.02) (Table 3). We identified six predictor measures that were individually related to improvements in Targets Hit. They included Extrafoveal Hits (\(\beta\)=0.70, \({f}^{2}\)=1.52, \(p\)<10^–6) (Fig. 7a), Objects Foveated (\(\beta\)=0.59, \({f}^{2}\)=0.80, \(p\)<10^–6) (Fig. 7b), Gaze-Hand Distance (\(\beta\)=0.58, \({f}^{2}\)=0.65, \(p\)<10^–6) (Fig. 7c), Mean Hand-Speed (\(\beta\)=0.50, \({f}^{2}\)=0.48, \(p\)<10^–6) (Fig. 7d), Target Foveation Time (\(\beta\)=-0.46, \({f}^{2}\)=0.41, \(p\)<10^–6), and Foveation Time Difference (\(\beta\)=0.23, \({f}^{2}\)=0.09, \(p\)<10^–6). We also identified six predictor measures that were individually related to improvements in Distractors Avoided. They included Gaze-Hand Distance (\(\beta\)=0.29, \({f}^{2}\)=0.25, \(p\)<10^–6), Target Foveation Time (\(\beta\)=− 0.11, \({f}^{2}\)=0.04, \(p\)<10^–3), Hand-Speed Bias (\(\beta\)=0.11, \({f}^{2}\)=0.03, \(p\)<10^–3), Extrafoveal Hits (\(\beta\)=0.11, \({f}^{2}\)=0.03, \(p\)<10^–3), Foveation Time Difference (\(\beta\)=− 0.09, \({f}^{2}\)=0.02, \(p\)<10^–3), and Objects Foveated (\(\beta\)=0.09, \({f}^{2}\)=0.02, \(p\)<0.01).

We subsequently tested our second hypothesis by using multiple regression to analyze the extent to which refinements on the preceding predictor measures were independently predictive of improvements in Target Hits and Distractors Avoided (i.e., at least a small semipartial effect size, \({f}^{2}\)≥0.02) (Table 4). Our multiple regression identified two measures of visual search (Extrafoveal Hits: \(\beta\)=0.54, \({f}^{2}\)=0.61, \(sp\)<10^–6; Objects Foveated: \(\beta\)=0.32, \({f}^{2}\)=0.16, \(sp\)<10^–6), one measure of eye-hand coordination (Gaze-Hand Distance: \(\beta\)=0.22, \({f}^{2}\)=0.07, \(sp\)<10^–3), and one measure of skilled limb movement (Mean Hand-Speed: \(\beta\)=0.14, \({f}^{2}\)=0.03, \(sp\)=0.02) that were independently predictive of improvements in Target Hits (Fig. 8a). In contrast, our multiple regression only identified a single measure of eye-hand coordination (Gaze-Hand Distance: (\(\beta\)=0.24, \({f}^{2}\)=0.04, \(sp\)=0.01) that was independently predictive of improvements on Distractors Avoided (Fig. 8b).

Finally, our stepwise regression analyses confirmed the results obtained from our multiple regression analyses. Specifically, the final model for Targets Hit only included the same measures of visual search (Extrafoveal Hits, Objects Foveated), eye-hand coordination (Gaze-Hand Distance) and skilled limb movement (Mean Hand-Speed). Furthermore, the final model for Distractors Avoided only included Gaze-Hand Distance.

The results of this study provide indirect evidence that practice-related refinements involving multiple neural processes may contribute to motor learning. Notably, we observed that measures of skilled limb movement, visual search and eye-hand coordination underwent practice-related refinements (Hypothesis 1) that were independently predictive of improvements in task performance (Hypothesis 2). Importantly, in drawing this conclusion, we assume that the trial-by-trial and week-by-week refinements exhibited by measures of skilled limb movement, visual search and eye-hand coordination can be used to infer that practice produced refinements involving multiple neural processes. Furthermore, we assume that motor learning could be inferred from trial-by-trial and week-by-week improvements exhibited by measures of task performance.

Other studies have provided evidence that both sensory and motor processes contribute to motor learning [48], but these studies were not designed to investigate the extent to which these processes are independent predictors of motor learning. As result, we do not know the extent to which relationships with motor learning reflected independent or shared variance. In the current study, we addressed the issue of covariation by examining independent predictions of motor learning after removing all shared variance. This analysis showed that skilled limb movements, visual search and eye-hand coordination are independent predictors of motor learning, indicating that studies of motor learning should account for the various processes that may influence improvements in task performance.

Increases in Mean Hand-Speed were associated with increases in Targets Hit, indicating that participants learned to hit more targets by quickly moving their hands to different areas of the workspace. Although faster movements are more variable and less accurate [2, 9], any decreases in movement accuracy were not associated with increases in the proportion of hand movements that failed to make paddle-contact with targets. Alternatively, it is possible that optimization of intermuscular coordination may have allowed participants to move faster without incurring greater movement variability. In either case, increases in movement speed had a positive effect on task performance, thus our results are consistent with the principles of optimal feedback control [49, 50].

Increases in Extrafoveal Hits and Objects Foveated were the strongest independent predictors of increases in Targets Hits. These findings indicate that refinements of visual search led to better task performance by optimizing how participants gathered information with foveal and extrafoveal vision. This is consistent with evidence that visual search is highly adaptive to different task demands and environments, such as environments in which task-relevant objects are more likely to appear at certain locations [51, 52].

The association between Extrafoveal Hits and Target Hits indicates that participants learned to use extrafoveal information to guide hand movements used to hit targets. This is consistent with a previous study of visual search, which found that practice led to improvements in using extrafoveal vision to search for objects with task-relevant features [53]. In addition, cortical areas known to process peripheral visual information exhibit greater involvement during motor tasks [54]. However, to our knowledge, our study is the first to show that refinements of extrafoveal visual processing are predictive of motor learning.

The association between Targets Foveated and Target Hits suggests that refinements used to maximize the number of objects that participants foveated with visual search led to improvements in hitting targets. The modest correlation between Objects Foveated and Target Foveation Time (r = − 0.31; Table 1) also indicates that, at least in part, decreases in the time spent foveating targets freed up time to foveate more objects. In contrast, studies of “quiet eye” have found that experts at motor tasks have longer foveations on task-relevant objects than novices [17, 19, 20]. Furthermore, training interventions designed to increase foveation durations have produced improvements in motor performance [22‐25]. These divergent findings suggest that both increases and decreases in foveation times can benefit motor performance, depending on the task demands and environment. As a result, we predict that practice will lead to increases in target foveation times in tasks with high demands on accuracy and low demands on speed of visual processing, whereas practice will produce decreases in foveation times in tasks with low demands on accuracy and high demands on speed of visual processing.

Increases in Gaze-Hand Distance were associated with increases in Targets Hits, indicating that looking away from targets before hitting them led to improvements in task performance. Although this contrasts with studies showing rigid coupling between initiation of eye movements and completion of hand movements [43], other studies have found that this rigid coupling decreases with practice [33‐36]. We believe that increases in Gaze-Hand Distance may reflect a transition from an early reliance on visual feedback for accurate execution of hand movements to a subsequent reliance on kinesthetic feedback for accurate execution of hand movements. This would have allowed visual search to gather task-relevant information with greater efficiency [33]. Specifically, looking away from targets before hitting them would have disrupted visual feedback used to accurately guide hand movements toward targets. However, it would have enabled earlier and longer foveations of objects, thereby facilitating more efficient decisions whether to hit or avoid objects by either executing or inhibiting skilled limb movements. Importantly, any negative effects resulting from disruption of visual feedback of hand movements could be offset by a greater reliance on kinesthetic feedback, which is known to improve during motor learning [55‐57] and may directly contribute to motor learning [58‐61].

We found that motor execution (Targets Hit) and motor inhibition (Distractors Avoided) exhibited distinct patterns of improvements. Notably, Targets Hit showed trial-by-trial and week-by-week improvements, whereas Distractors Avoided displayed only week-by-week improvements. We also found that different processes were independently predictive of improvements in motor execution and inhibition. Refinements of skilled limb movements (Mean Hand-Speed), visual search (Objects Foveated, Extrafoveal Hits) and eye-hand coordination (Gaze-Hand Distance) were independently predictive of improvements in Targets Hit. In contrast, eye-hand coordination (Gaze-Hand Distance) was the only independent predictor of improvements in Distractors Avoided. Given that avoiding distractors mainly involved inhibition rather than execution of hand movements, it is not surprising that increases in Mean Hand-Speed were not predictive of increases in Distractors Avoided. In contrast, increased Gaze-Hand Distance would have facilitated both motor execution and inhibition by allowing participants more time to make decisions whether to initiate or inhibit movements. It is perhaps surprising that increases in Objects Foveated were not predictive of increases in Distractors Avoided. We would expect that more efficient visual search should lead to improvements in both motor execution and inhibition by allowing more objects to be processed with foveal vision. The lack of a relationship may reflect that participants exhibited smaller improvements on Distractors Avoided. However, if the proportion of targets and distractors was equal or reversed, we expect that our participants may have shown greater improvements on Distractors Avoided and we may have found a meaningful relationship.

By examining patterns of variability exhibited by measures related to multiple neural processes, we found that refinements of multiple processes were independently predictive of motor learning. However, our paradigm and analyses were not designed to make causal inferences. This requires measuring motor learning while experimentally manipulating one process and controlling for interactions with all other processes. For example, masking objects that are not located within foveal vision would neutralize the contributions of extrafoveal hits on motor learning. If this reduced motor learning without affecting refinements of other processes, it would show that refinements of extrafoveal processing are causally linked to motor learning.

Another limitation of the current study is that we did not examine practice-related refinements of proprioception. This is an important limitation because improvements in planning and executing skilled limb movements may involve refinements that alter the processing of proprioceptive feedback [37]. In agreement with this hypothesis, previous studies have demonstrated that motor learning is associated with modifications of rapid responses to proprioceptive feedback [62] and improvements in kinesthesia [54‐56]. Although we do not know if refinements involving proprioceptive processing contribute to motor learning in the current study, we believe they may have facilitated increases in Gaze-Hand Distance by reducing reliance on visual feedback used to accurately execute skilled limb movements.

Task demands and environmental features are known to alter motor learning [63, 64]. However, we did not investigate how task demands and environmental features influence the extent to which different processes are predictive of motor learning. In the current paradigm, for example, we would expect that refinements of skilled limb movements would be a greater predictor of motor learning if the demands on skilled limb movements were increased by reducing the size of the paddles or by imposing mechanical perturbations on the hands.

Although our behavioral measures probed several neural processes involved in motor learning, we did not directly investigate the underlying neural mechanisms of motor learning. Numerous studies of motor learning have explored changes in brain regions and networks related to refinements of skilled limb movement [65‐67]. Other studies have investigated the brain regions and networks associated with visual search during perceptual and cognitive tasks [68‐72]. However, we are unaware of any studies that have examined the extent to which brain regions and networks that underlie multiple processes are associated with motor learning.