Introduction

Cognitive abilities such as sensory perception, spatial localization, and working memory can be improved through extensive training on related tasks. For example, perceptual sensitivity can be dramatically improved after extensive perceptual learning in discriminating visual features1,2,3. However, such repetitive training on a specific task is usually boring. Recently, much effort has been spent on embedding cognitive training into video or sports games4,5. People found that playing video games can improve various cognitive abilities, including contrast sensitivity, attention, problem-solving, and memory6,7,8,9. In addition, extensive research has explored exercise’s general physical, cognitive, and mental benefits, such as decision making, problem-solving, stress relief, and the alleviation of symptoms associated with depression and anxiety, possibly by increasing blood flow, boosting neurotransmitter production, and promoting the growth of new neurons10,11,12,13,14.

However, previous research on the effects of sports training on cognitive function has mainly focused on high-level cognitive abilities, such as decision-making, attention, memory, and learning in adults. The extent to which sport training affects fundamental sensory coding and sensorimotor integration remains under-investigated. More importantly, it is still unclear the extent to which the cognitive benefit of a particular sports activity can transfer to a completely new context to enhance the new mapping between environment and cortical representation, especially in children, a crucial question for the understanding of sensorimotor development and the curriculum design in schools14.

There are several reasons to predict this transfer. First, the active vision framework emphasizes that visual processing primarily serves to guide actions, positing that we actively explore the visual world through movement to accomplish behavioral goals15. Within this framework, the precise, goal-directed actions inherent in sports are theorized to fine-tune visual systems. Second, a key tenet of body schema is that the brain’s representation of the body is dynamic, capable of incorporating tools as extensions of the body schema16,17. From this view, proficient sports abilities may not only constitute a motor skill but could fundamentally recalibrate the user’s sensorimotor parameters and spatial mapping. Third, the predictive processing theory proposes that the brain continuously refines its internal models, including perception, action, and cognition, through sensorimotor experience to better predict its sensory inputs and the sensory consequences of action18,19. An internal model allows an organism to simulate the consequences of potential actions without physically executing them. This predictive capability is crucial given the inherent neural delays in processing sensory information. Based on these models, it seems reasonable to predict that what one learns in the sports field should modify the fundamental internal model and transfer to new contexts.

In fact, previous studies have shown that adult athletes have better spatial localization abilities than nonathletes when asked to blindly walk to a previewed object20, or to verbally estimate the distance to an object21,22, suggesting the generalization of sports training to new contexts. However, one limitation of these studies is that they tested only adult athletes, leaving open the question of whether the same benefit exists for children with shorter durations of sport training, a question of great interest to parents and education policymakers. Moreover, previous studies used standard integer distances (e.g., 10 m in 20; 2 to 9 m at 1-m intervals in 21). Participants may see marks of these distances in life, for example, on floors in stores, banks, and airports to space people in line. Students may see these marks in math and gym classes, as well as in running and jumping tests. It is still unclear whether students with tennis training and adult tennis athletes can perform better even at random, non-integer distances (e.g., 3.15 m, 6.85 m, 11.65 m) that were rarely marked in any aspect (accuracy, noise, or scaling). Moreover, the random, non-integer distances help to force participants to rely on the immediate integration of visual input and proprioception.

To examine the generalization of sport training to spatial abilities in purely unfamiliar contexts, we tested blind walking performance in sighted children and adults who received tennis training. We selected tennis as an exemplar sport to address our research question because during tennis training, players need to view, approach, and hit the ball in precise distances, a process involving continuous visual monitoring, prediction of externally defined spatial targets, and precise calibration between visual distance and whole-body action. Blind walking was selected as the testing task because it is a widely used paradigm in psychological research to assess action based on distances (distance-guided action)23,24,25,26,27,28. It involves having participants observe and memorize an object’s location, and then attempt to walk to the remembered spot while blindfolded. Rieser et al.24 evaluated adults’ absolute egocentric distance perception accuracy within 24 meters in open fields, highlighting the essential role of proprioceptive cues in blind walking calibration. Ooi et al.23 and Wu et al.28 extended and officially proposed the blind walking paradigm, displaying the human visual system’s reliance on surface integration for accurate distance judgment. Philbeck and Loomis29 advocate action-based approaches, such as walking or touching targets, for egocentric distance perception. In summary, the blind walking paradigm requires integrating visual and proprioceptive information to form internal sensorimotor representations and is a reliable and effective experimental method for studying human spatial cognition27.

Tennis players were trained in the light in an open field of a constant size, but during the blind-walking test, they viewed a target, were then blindfolded, and walked to it at unfamiliar distances, with each distance tested only once. Therefore, the blind walking task is in a completely different context for them and requires integrating remembered visual information with proprioception, which is only available during that trial. Any improvement in the blind-walking performance of tennis players relative to controls would provide evidence of a general benefit of tennis training.

Materials and methods

Participants

Seventy-six participants were recruited for this study. Five of them were excluded for walking out of the experimental fields or for having a walking distance that deviated by more than 3 standard deviations from the mean on more than one trial, leaving a total of 71 participants for analysis. Experiment 1 examined the effects of tennis training on children’s blind-walking performance. There were two groups of participants. The tennis group comprised 21 children (9.62 ± 1.66 years old, mean ± SD; 15 females; 6 males) with tennis training experience (training duration: 2.27 ± 1.33 years, mean ± SD). The control group comprised 22 children (9.09 ± 1.97 years old, mean ± SD; 9 females; 13 males) with no prior training in tennis or other ball sports, but all participated in extracurricular physical activities such as dance, swimming, skipping, and fitness training for similar number of hours as the tennis group (typically, 1–8 h per person per week). The sports activities of the control group do not require a spatial scale movement as large as a half tennis court (e.g., skipping and fitness in a fixed spot or dancing with over 5 people in a room as large as a half tennis court) or do not require walking movements on the ground (e.g., swimming). The control group all participated in more than 2 sports and changed sports almost every term to try different activities before testing. Therefore, they did not have a fixed primary sport.

Similarly, two groups of adult participants were recruited for Experiment 2, which examined the effects of over 5 years of tennis training on blind walking performance at unfamiliar non-integer distances. The tennis group consisted of 13 professional tennis players (21.85 ± 4.22 years old, mean ± SD; all males, as the coaches through whom the participants were recruited worked only with male players) with extensive professional tennis training (training duration: 10.77 ± 5.12 years, mean ± SD). Eleven of them met the Chinese national second- or first-class athlete levels (achieved high ranking in designated national or international matches), while the remaining two were coaches at a tennis club. The control group comprised 15 participants (23.87 ± 3.16 years old, mean ± SD; all males) without prior tennis or other ball sports training and all engaged in ball sports games less than once per month for at least 2 years before the test.

According to G*Power (https://www.psychologie.hhu.de/)30,31, to achieve adequate statistical power to detect the effect of the between-group factor (i.e., 2 Groups) with 4 measures (4 Distances) using a mixed-design ANOVA, the total sample size should be at least 36 (Effect size = 0.25, alpha = 0.05, Power = 0.95), which is smaller than our sample size (N = 43) in children. Moreover, the G*Power results indicate that to achieve adequate power to detect the effect of the between-group factor (i.e., 2 Groups) with 7 measures (7 Distances) using a mixed-design ANOVA, the total sample size should be at least 26 (Effect size = 0.25, alpha = 0.05, Power = 0.95), which is also smaller than our sample size (N = 28) in adults. These results suggest that our study had adequate power to detect the effects we expected.

Participants in the tennis groups (both Experiment 1 and Experiment 2) were recruited through a tennis coach. The child control group participants (Experiment 1) were recruited from the same primary schools as the tennis groups indicating that they had similar living environments, physical activity levels, socio-economic status, and general cognitive abilities. The adult control group was recruited from South China Normal University through posters. Because the adult control group and the adult tennis players were all from the tier 1 universities in China, it is reasonable to assume that they had similar general cognitive abilities. However, the physical activity levels of them were apparently distinct between the athletes and the control group.

All participants had normal or corrected-to-normal vision and were unaware of the purpose of the experiments. They all received compensation for their participation. Informed written consent was obtained from adult participants and the parents or guardians of the child participants. The study was approved by the ethics committee of South China Normal University (SCNU-PSY-2019-4-33), and the methods were in accordance with the guidelines established in the Declaration of Helsinki.

Apparatus and stimuli

In Experiment 1, 15 children from the tennis group completed the experiment in an open field at one end of a tennis court. This field featured a flat, unmarked wall approximately 15 m long, well beyond the farthest distance tested for children (i.e., 6.85 m). A one-meter-wide, twenty-meter-long red carpet was laid out on the field to eliminate ground-based distance cues. Due to variant restrictions in booking the tennis court, the remaining participants, including the rest of the six children from the tennis group and all children in the control group, completed the experiment in a corridor of a building at South China Normal University with the same setup including the same red carpet, aiming to match the testing environment (Fig. 1A). The pilot analysis showed that the testing locations did not affect the error or the correlation results (Table S1), and therefore the data were pooled together in the following analysis. In Experiment 2, all adult participants completed the experiment in the same corridor with the same setup.

Participants wore opaque goggles to eliminate visual information during the blind walking task. A laser rangefinder (Deli, DL331040D; precision, ± 3 mm) was employed to measure the walking distances accurately. The experimental setup is shown in Fig. 1A. A paper board was used to mark the starting position and aid in measuring distance. Participants walked while blindfolded along the red carpet. They were instructed to touch the flat wall near the tennis court or the smooth metal beam while walking in the corridor to restrict walking direction and avoid veering left or right. The wall and the metal beam were smooth and could not provide tactile cues. A toy cone indicated the target with its closer edge as the reference.

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

Experimental setup and illustration of signed error and absolute error. (A) Participants walked blindfolded with the assistance of a smooth beam in the corridor (or a smooth wall) to guide the walking direction. One side of a paper board was aligned with the participant’s tiptoe. The physical target distances were measured from the cone to the board (the green distance arrow). After reaching the target position, the walked distance was measured from the participant’s tiptoe to the board. Experiment 1 involved child participants with four relatively short target distances (1.55 m, 3.15 m, 4.95 m, and 6.85 m). Experiment 2 involved adult participants with seven target distances (3.15 m, 4.85 m, 6.35 m, 8.75 m, 10.15 m, 11.65 m, and 13.65 m). These distances approximately follow a uniform distribution and are selected randomly to prevent participants from making an estimation based on their prior life experience. Each participant was tested only once for each distance, and no feedback was given. (B) The red triangles represent the walked distances of an example participant, and the grey circles represent the target distances. The grey dashed line represents the correct response, where the walked distances equal the target distances. Signed error (SgnE) was calculated as the walked distance minus the target distance, which could be positive or negative. A negative signed error indicated that the participant underestimated the target distance in the current trial (e.g., SgnE2) during walking. Absolute error (AbsE) was the absolute value of the SgnE, which was always positive and also known as unsigned error. The SgnE4 and AbsE4 were not depicted for conciseness.

Design and procedure

The procedures for Experiment 1 and Experiment 2 were identical, except that Experiment 1 comprised four distances, whereas Experiment 2 comprised seven. The order of the target distances was randomized. Both experiments required two experimenters: experimenter A stood near the starting point, and experimenter B stood near the target location.

All participants had a short practice session to familiarize themselves with the task before the experiment, with distances other than the distances that would be tested. This included understanding the procedure, training to rapidly remove and put on the goggles, and practicing natural blind-walking gaits. The practice session lasted about 5 min.

At the beginning of each trial, participants were blindfolded and stood at a random starting point, and the toy cone was placed at the target position of the current trial. Participants were instructed to take off the goggles and view the target for about one second, then put the goggles back on and walk blindfolded up to the target. The toy cone was removed as soon as the participants walked. In order to exclude any distance cues from experimenters, both of them remained stationary outside the red carpet and kept silent when participants were walking. After measuring the walked distance, the participants, who kept wearing the goggles, were guided back to a random starting point for the subsequent trial. Participants were only allowed to take off their goggles when viewing the toy cone at the beginning of each trial. They did not receive any feedback about their walking distance. Each distance was tested only once.

Data analysis

The primary dependent variable was the walked distance, defined as the distance walked by participants in each trial. To evaluate the performance of blind walking, we calculated the signed error (SgnE) and absolute error (AbsE) at each target distance condition for each participant. The signed error was calculated as the difference between the walked distance and the target distance, which could be positive or negative. Signed error reflected the bias and systematic deviations in performance, such as overshooting or undershooting a target, and could be addressed through explicit feedback in error-based learning32. The absolute unsigned error was calculated as the absolute difference between the walked distance and the target distance, which is always positive. Absolute errors reflected the bias and noise of the walked distance, representing random variability in performance, and can be enhanced through reward or loss in reinforcement learning.

They can be expressed as follows:

$${\text{S}\text{g}\text{n}\text{E}}_{i}={\text{W}\text{D}}_{i}-{\text{T}\text{D}}_{i}$$
$${\text{A}\text{b}\text{s}\text{E}}_{i}=\left|{\text{W}\text{D}}_{i}-{\text{T}\text{D}}_{i}\right|$$

Where SgnE is the signed error, AbsE is the absolute error, WD is the walked distance, TD is the target distance, and i is the target distance condition (Fig. 1B).

Activities like tennis, which require precise visual-spatial coordination and motor execution, may enhance perceptual-motor integration and consistency, thereby reducing individuals’ bias and noise. Therefore, we predicted that both SgnE and AbsE would be improved by tennis, especially in children.

Because the Shapiro-Wilk test revealed that the SgnE or AbsE data at some distances do not follow the normal distribution, nonparametric statistical analyses were performed. Specifically, the Wilcoxon signed-rank test was performed to compare SgnE and AbsE against zero for each distance within each group in both experiments. Rank-Biserial Correlation was reported as the effect size for the Wilcoxon signed-rank test.

The two-way mixed-design Aligned Rank Transform (ART) ANOVA33,34 with Groups (tennis group vs. control group) as a between-subject factor and Distances (four target distance conditions in Experiment 1 and seven target distance conditions in Experiment 2) as a within-subject factor to examine the difference between the two groups in SgnE and AbsE. ART procedure enables nonparametric factorial ANOVA while preserving the ability to test main effects and interactions in mixed designs34. Significant interactions were followed up with post hoc pairwise comparisons within the ART procedure33. Multiple comparisons were corrected using Holm’s sequential Bonferroni procedure35. Partial eta squared (ηp²) was reported for effect size in ART ANOVA. All ART analyses were conducted with the ARTool package (http://depts.washington.edu/acelab/proj/art/) in R (https://www.r-project.org/).

Importantly, in Experiment 1, to remove the possible effect of gender, we first included gender as a covariate to regress out the contribution of gender in the SgnE and AbsE data using the regstats function in MATLAB2023a (https://www.mathworks.com/) before performing the ART ANOVA. No such adjustment was made in Experiment 2 because all participants in both groups were male.

In addition to the signed and absolute errors, we also calculated the Pearson’s correlation coefficients between the walked distances and target distances, following the methods used by Brungart36 and Kopco & Shinn-Cunningham37 to evaluate the extent to which participants could blindly walk to the prior target. Higher coefficients indicate a stronger relationship between the walked and target distance, i.e., a more consistent sensorimotor scaling, which provides a further assessment of the blind walking ability according to the target distance.

Moreover, because it has been suggested that the relationship between walked distances and target distances is not linear for each subject and distance, and therefore should be transformed to logarithms before calculating Pearson’s correlation38, we also transformed the walked distances into logarithms to calculate Pearson’s correlation on a log-log scale36,37.

All Pearson’s correlation coefficients were Fisher z-transformed to satisfy normality assumptions. As with SgnE and AbsE, the Fisher Z scores in Experiment 1 also underwent preprocessing to remove the possible effect of gender. No such adjustment was made in Experiment 2 because all participants were male. The Shapiro–Wilk test confirmed that the residuals of the Fisher Z scores were normal on both the raw and log-log scales. Therefore, the residuals were carried forward to subsequent independent-samples t-tests in JASP (https://jasp-stats.org/ )39 to compare the correlation coefficients between Groups. Cohen’s d was used as an effect size for the independent-samples t tests.

Results

Experiment 1: children

In Experiment 1, we compared the walking distance performance of the tennis group children with that of the control group of children across four target distances (1.55 m, 3.15 m, 4.95 m, and 6.85 m). Figure 2A presents the walked distances across participants as a function of target distances for each individual. Visual inspection reveals that the mean walked distances (opaque lines) for both groups were close to the correct response (grey dashed line), implying minimal systematic bias in children. This observation is supported by our statistical results (Fig. 2B): both the tennis group and the control group showed no significant signed errors in any target distance conditions (Wilcoxon signed-rank test; all pcorrected > 0.120), suggesting that both groups exhibited no bias in walked distance (see Table S1 for details). ART ANOVA showed that the main effects of Group (F(1,41) = 0.675, p = 0.416, ηp² = 0.016), Distance (F(3,123) = 0.305, p = 0.822, ηp² = 0.007) or their interaction (F(3,123) = 0.813, p = 0.489, ηp² = 0.019) were not significant, suggesting that there was no difference in bias across participants or across distances.

Next, we analyzed the absolute error results. Given that no significant bias was observed for the two groups, the absolute error indicated the noise of distance performance across participants. For absolute error, both the tennis group and the control group demonstrated significant absolute error in all target distance conditions (Wilcoxon signed-rank test; all pcorrected < 0.001, all Rank-Biserial Correlation = 1.000, Fig. 2C). ART ANOVA showed that the interaction between Group and Distance (F(3,123) = 1.069, p = 0.365, ηp² = 0.025) and the main effect of Distance (F(3,123) = 0.865, p = 0.461, ηp² = 0.021) were not significant, but the main effect of Group (F(1,41) = 4.700, p = 0.036, ηp² = 0.103) was significant. The absolute error of the tennis group was lower than that of the control group (see Table S2 for details). These results suggested that children with tennis training show less noise in blind walking (Fig. 2C).

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Results of signed error and absolute error in Experiment 1. (A) Line plots of walked distances of the tennis group (N = 21, red lines) and control group (N = 22, grey lines) as a function of target distances. Semitransparent lines represent individual-level data. The grey dashed line indicates the ideal response, where walked distances equal target distances. (B) Signed error on four target distance conditions for both the tennis and control groups. The red circles represent the mean for the tennis group, and the grey circles represent the mean for the control group. (C) Absolute error on the four distance conditions for both the tennis and control groups. The red circles represent the mean for the tennis group, and the grey circles represent the mean for the control group. The asterisks above the x-axes indicate the difference between AbsE and zero under the corresponding distance conditions. The asterisks in the upper left corner indicate the significance of the main effect of Group. * denotes that the main effect of Group was significant at p < 0.05, *** denotes that the absolute error was significantly larger than 0 at pcorrected < 0.001, Holm correction. Error bars depict ± 1 standard error.

In addition to the signed and absolute errors, we also calculated the Fisher z-transformed Pearson’s correlation coefficient (Fisher Z score) between the walked distances and target distances. The correlation reflected how well participants scaled their walked distances to the target distances36,37,40,41. The possible effect of gender was removed from the Fisher Z score before comparison across groups. The tennis group’s Fisher Z score was higher than that of the control group (Fig. 3B; t(41) = 2.888, p = 0.001, Cohen’s d = 1.077).

We also transformed the distances to logarithms (Fig. 3A) and calculated the correlation coefficients on a log-log scale (Fig. 3C) because it has been suggested that the relationship between walked distances and target distances is not linear for each subject and distance, and therefore should be transformed to logarithms before calculating Pearson correlation38. The tennis group still performed better than the control group (t(41) = 2.805, p < 0.001, Cohen’s d = 1.121). Overall, these results suggested that tennis training is associated with reduced sensorimotor noise and more consistent distance scaling during blind walking in school-age children.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Results of correlation between walked distance and target distance in Experiment 1. (A) Line plot of the logarithm-transformed walked (lg-walked) distances for the tennis group (N = 21, red line) and control group (N = 22, grey line) as a function of the logarithm-transformed target (lg-target) distances. Semitransparent lines represent individual-level data, which are more linear than raw scale data and approximately linear after logarithmic transformation. The grey dashed line indicates the ideal response, where walked distances equal target distances. (B and C) Fisher Z scores of the correlation coefficients between walked and target distance in the raw scale (B) and the logarithmic scale (C). The thick red lines represent the mean for the tennis group, and the thick grey lines represent the mean for the control group. Dots represent individual-level data. Note that data preprocessing was conducted to remove the effect of gender on Fisher Z or Fisher Z (log) before the statistical comparison. However, the raw Fisher Z and Fisher Z (log) data were shown in the figure for intuitive representation. ** denotes p < 0.01, *** denotes p < 0.001. Error bars depict ± 1 standard error.

Experiment 2: adults

In Experiment 2, we compared the walking distance performance between the tennis and control groups among adult participants. The analyses were exactly the same as those in Experiment 1. Figure 4A presents the walked distances across participants as a function of target distances. Both groups had no significant signed errors (Fig. 4B) in any target distance conditions (Wilcoxon signed-rank test; all pcorrected > 0.119), suggesting that both groups exhibited no bias in walking distance. ART ANOVA showed that the main effects of Group (F(1,26) = 0.719, p = 0.404, ηp² = 0.027), Distance (F(6,156) = 1.778, p = 0.107, ηp² = 0.064) or the interaction (F(6,156) = 1.796, p = 0.103, ηp² = 0.065) between Group and Distance were not significant. These results suggest that similar with children, there was no significant bias for both groups of participants.

Both the tennis group and the control group demonstrated significant absolute errors for all target distance conditions (Wilcoxon signed-rank test; all pcorrected < 0.001, all Rank-Biserial Correlation = 1.000, Fig. 4C). ART ANOVA showed that the main effect of Group (F(1,26) = 1.423, p = 0.244, ηp² = 0.052) was not significant, but the main effect of Distance (F(6,156) = 18.595, p < 0.001, ηp² = 0.417) was significant. The AbsE of the participants increased with the Target distance. The interaction between Group and Distance (F(6,156) = 2.824, p = 0.012, ηp² = 0.098) was also significant. Post hoc analysis revealed that before correction, there was a significant difference between the two groups of AbsE at a distance of 11.65 m (t(86.64) = 2.52, puncorrected = 0.014, Cohen’s d = 0.269). However, after correction, there was no difference between the groups at any distance condition (all pcorrected > 0.095).

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

Results of signed error and absolute error in Experiment 2. (A) Line plot of walked distances for the tennis group (N = 13, red line) and control group (N = 15, grey line) as a function of target distances. Semitransparent lines represent individual-level data. The grey dashed line indicates the correct response, where the walked distances equal the target distances. (B) Signed error in the seven target distance conditions for both the tennis and control groups. The red circles represent the mean for the tennis group, and the grey circles represent the mean for the control group. (C) Absolute error in the seven target distance conditions for both the tennis and control groups. The red circles represent the mean for the tennis group, and the grey circles represent the mean for the control group. The asterisks above the x-axes indicate the difference between AbsE and zero under the corresponding distance conditions. The asterisks in the upper left corner indicate the significance of the main effect of Distance. Red stars indicate how significant the difference between the tennis group and 0 is, while grey stars show how significant the difference between the control group and 0 is. ** denotes pcorrected < 0.01, *** denotes pcorrected < 0.001, Holm correction. Error bars represent ± 1 standard error.

Consistent with Experiment 1, we also calculated the Fisher Z score of the correlation between the target distance and walked distance for both the tennis and control groups. The results revealed that the Fisher Z score of the tennis group was higher than that of the control group (Fig. 5B; t(26) = 2.104, p = 0.045, Cohen’s d = 0.797), suggesting that more consistent sensorimotor scaling between target and walked distances observed in children was also observed in adults. We also transformed the distances to logarithms (Fig. 5A) and calculated the Fisher Z score on a logarithmic scale (Fig. 5C). The tennis group still showed more consistent scaling than the control group (t(26) = 2.314, p = 0.029, Cohen’s d = 0.416). These results suggested that the adult tennis group can better scale their walked distance with their target distance than the control group.

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Results of Fisher Z scores in Experiment 2. (A) Line plot of mean lg-walked distances for the tennis group (N = 13, red line) and control group (N = 15, grey line) as a function of lg-target distances. Semitransparent lines represent individual-level data, which are more linear than original scale data and approximately linear after logarithmic transformation. The grey dashed line indicates the correct response, where the walked distances equal the target distances. (B and C) Fisher Z scores were calculated from distances in the original scale (B) and logarithmic scale (C). The thick red line represents the mean for the tennis group, and the thick grey line represents the mean for the control group. Dots represent individual-level data. Error bars represent ± 1 standard error. * denotes p < 0.05.

It is notable that the height of the tennis group was significantly higher than that of the control group (Mann-Whitney U test; U = 154.500, p = 0.009) in Experiment 2. This indicated that the participants in the tennis group faced a larger deviation between observed distances (i.e., the distance between eyes and targets) and target distances (i.e., the horizontal distance between starting points and targets). Previous studies revealed that the higher the eye height was, the poorer the performance in distance estimation would become42,43, which indicated that taller participants had more difficulty in estimating distances accurately. Even in this case, the tennis group showed a higher correlation between walked distance and target distance, suggesting that tennis training may achieve even more consistent scaling if their eye height is matched with that of the control group.

Discussion

This study examined the relationship between tennis training and spatial cognition in real-world environments among both children and adults. We found that training in tennis can improve walking performance in some aspects in a completely new context. Specifically, even though participants were trained to approach the target ball using visual information and online visual feedback while playing tennis, they showed improvement in approaching the target position when blindfolded and walking to it based on remembered visual information at unfamiliar distances. This finding suggests that the influence of sports training on cognitive spatial ability is not specific to the context but could be generalized to new contexts. Given that navigating without sight is a common everyday experience—such as walking to bed after turning off the lights—our findings also suggest that field-based sports training may improve practical life skills.

The more consistent sensorimotor scaling observed in tennis players in the blind walking task is particularly revealing given the contextual opposition between the blind walking task and their training environment – that is, blind walking in a blindfolded state while tennis training with full viewing in light. We propose that this generalization is driven by common computational demands shared between the training and testing contexts, specifically distance-dependent motor control—the ability to precisely map visual distance cues to motor outputs44. Consequently, our results suggest that tennis training refines this predictive scaling of space, calibrating a more fundamental and stable internal model of egocentric space45. This interpretation is consistent with the predictive processing framework, which posits that skilled performance relies on the brain’s ability to generate accurate predictions of sensory consequences.

Importantly, the improvement was also observed in children who were only trained for a short duration, ranging from 1 to about 3 years, which goes beyond previous studies that show adult athletes perform better on blind walking or distance estimation20,21,22 than nonathletes. The blind walking task requires accurate vestibular and proprioceptive information44 and the ability to integrate visual and proprioception information with motor spaces to establish internal representations according to the internal model hypothesis. However, the internal sensorimotor representations established in children are often inaccurate and require refinement through natural development and deliberate practice45. This deficiency is reflected in children’s unsmooth and unstable performance in various motor tasks, such as aiming46, force production47, and grasping48. Accordingly, while children in the control group exhibited the expected high variability and lower sensorimotor consistency, tennis-trained children demonstrated significantly reduced absolute errors and stronger correlations. This suggests that the advantage associated with tennis training in childhood reflects not only refined sensorimotor scaling but also a distinct reduction in processing noise. This implies that tennis training may facilitate the establishment of more stable internal models during the critical period of sensorimotor development.

In addition to changes in the internal model, our findings also align with the concepts of “active vision” and “embodied cognition,” underscoring the importance of active movements in spatial cognition. These frameworks collectively provide a compelling basis for why tennis-training accompanied refinements should transfer to a novel, non-sport-specific task like blind walking. The active vision framework posits that visual encoding is optimized through the demands of goal-directed movement. Tennis training constitutes a highly rigorous regime of such active vision, forcing the brain to continuously solve the problem of “how do I get my body (and racket) to the right place at the right time?” This constant calibration of vision to action in a rich, dynamic environment appears to produce a more generalized spatial acuity that is no longer tethered solely to the sport’s context.

Furthermore, this transfer is also predicted by an embodied cognition perspective, enriched by the predictive processing framework. Through thousands of repetitions, tennis players do not just learn a set of context-bound motor commands; they develop a refined internal model of their body in action within a spatial world. This model incorporates the racket as a functional extension of the body schema, fundamentally recalibrating the perceptual-motor system’s parameters for judging effective interaction distances. Blind walking, devoid of a racket or a moving ball, taps into this same core competency: the ability to generate an accurate motor plan from a remembered visual scene. The brain’s improved capacity to predict the sensory consequences of self-motion—a skill honed on the tennis court—directly supports the feedforward execution of the blind walking task. Therefore, the observed generalization is not merely a transfer of a specific skill, but evidence of a fundamental enhancement of the central predictive models that govern spatial perception and action.

One may argue that the tennis group may only have improvement in distance perception, but not in sensorimotor integration. Here, we would like to highlight that motor execution relies on sensory information rather than the perception of sensory information. Previous studies reported that even when participants showed distortion in distance perception, they could still walk to the target accurately, demonstrating a dissociation between distance perception and walked distance49,50. This dissociation aligns with the theoretical framework of the two-visual-stream theory (i.e., the ventral visual stream mediating visual perception and the dorsal visual stream mediating the visually guided action)51,52, which is supported by patients who were not able to consciously report the perceived size or shape of objects, but could adjust their grasping posture according to their size and shape. Because distance perception and walked distance are dissociable, the effects of tennis training on these measures are independent research questions. The current study has focused on the effect on blind walking, an action task. Future studies can examine the effect of tennis training on distance perception in school-age children.

It is important to note that although we tested only tennis players, other sports that require precise walking distances may also yield similar beneficial effects. For example, Durgin & Li showed that adult golfers had better spatial judgements than nongolfers21. However, sports that prioritize proprioceptive precision within a small space (e.g., dance that only requires short position changes) or locomotion in stable environments (e.g., skipping or fitness training) may not induce the same sensorimotor transfer. This distinction is supported by the fact that children in the control group had similar physical activity levels to the tennis group. Yet, those participating in sports that do not require calibrating distance-dependent gain control relative to external targets did not show the refined sensorimotor scaling and reduced processing noise observed in the tennis group. That also suggests it is unlikely that the difference in blind walking between the adult athletes and controls was due to differences in general physical activity. Future research can test this possibility.

A consideration of the current results is the potential role of self-selection bias. As participants in this study were not randomly assigned to tennis training, the differences observed could be attributed to pre-existing individual differences (e.g., innate visuospatial ability or motor coordination) that influenced both the decision to play tennis and performance in our task, or to a combination of these factors. While the consistency of our findings across both children and adults lends weight to the interpretation of a training effect, and theories of embodied cognition provide a plausible mechanism for such an effect, we cannot rule out the contribution of pre-existing differences. Future longitudinal studies following individuals before and after they begin sports training are needed to conclusively establish a causal relationship.

Another limitation of the current study is that, although we tested both children and adults, we cannot draw strong conclusions about development because the sample sizes and target distances differed between the two age groups, and the differences in their blind walking performance may reflect neurocognitive maturation rather than tennis-specific training. A third limitation is that, in the adult group, only male participants were included due to practical constraints. Although it seems reasonable to speculate that similar effects would be observed for females, research data are needed, considering the documented gender differences in spatial orientation44.

Overall, our study has demonstrated that training in tennis can improve sensorimotor ability in a completely different context and affect sensorimotor ability in everyday life, even for children trained for around 2 years on average. These findings align with the concepts of “active vision” and “embodied cognition,” which underscore the importance of active motion in perception and cognition. Importantly, these findings have significant implications for educational policy and sports training, providing valuable insights into sensorimotor development and the potential benefits of incorporating sports activities into educational curricula.