Extrinsic Variability but not Intrinsic Variability Affects the Search for Synergies in De-novo Motor Learning

Extrinsic Variability but not Intrinsic Variability Affects the Search for Synergies in De-novo Motor Learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Extrinsic Variability but not Intrinsic Variability Affects the Search for Synergies in De-novo Motor Learning Anadi Mehta, Joanne Smith, David Travieso, Raoul M. Bongers This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6865411/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 22 Oct, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract De-novo motor learning involves emergence of novel task specific joint angle coordination patterns, called synergies, through structured change in joint angle variability, called search. Search implies variability in joint angles and task performance, but the understanding of the relationship between variability and learning is still unclear. This study examines the contributions of extrinsic variability - differing practice schedules in the task (blocked vs random) - and intrinsic variability - flexibility in individual’s movements – on learning, search and synergy formation. Participants were first categorized based on their intrinsic variability (i.e., joint angle covariation in an adaptive reaching task). Learning was then evaluated on a virtual lateral interception task with a redundant mapping between joint angles and virtual paddle position that necessitated forming new synergies. Overall, search reduced, and synergies emerged with practice. Intrinsic variability did not affect learning, search or synergy formation. Extrinsic variability affected search during practice, but did not interact with emergence of synergies. During practice, the uncontrolled manifold (solution sub-space within the joint space) was approached rapidly in the initial practice phases while the movement along the uncontrolled manifold was more gradual. We demonstrate search in joint space during learning is structured and extrinsic variability increases search behaviour. Biological sciences/Psychology Biological sciences/Psychology/Human behaviour Biological sciences/Neuroscience/Motor control Motor Learning Synergies Exploration Intrinsic Variability Practice Variability and Uncontrolled Manifold Analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Humans have the unique ability to increase their motor repertoire by learning complex movements in which the degrees of freedom in the neuromotor system are coordinated in novel patterns to perform new tasks. The abundant nature of the human movement system which provides more degrees of freedom (DoF) than are necessary for executing a certain action, is instrumental in learning these novel coordination patterns. For instance, one can intercept a ball at a certain location in 3D space through a multitude of movements by moving one’s arm along 7 different axes of rotations (DoF at the joint angle level of analysis) across 3 different joints (shoulder, elbow and wrist). These DoFs are believed to be organized as synergies that facilitate variability without compromising task execution. This movement variability can be regulated for learning new motor skills requiring novel synergies 1 , 2 . Our study is an attempt to identify how different sources of variability influence the search for new stable synergies when learning a de-novo task using the dynamical systems approach. Joint angle variability in performing actions can either be externally imposed through task manipulations (extrinsic) or can originate within a person (intrinsic). Extrinsic variability in movements is primarily studied by introducing differing practice conditions. Within the dynamical systems approach, variability of practice is considered to provide opportunities for exploration related to searching new movement solutions and stimulating more adaptive behavior 3 – 5 . A randomized practice schedule (high extrinsic variability) necessitates varied movements due to changing task conditions as compared to blocked practice (low extrinsic variability). Tuitert et al. 6 performed a reaching study with variable heights of obstacles and reported an increased range of motion in the joint angles but no change in covariation (i.e., compensatory variation) of the DoF within synergies during practice. This result may be attributed to the presence of stable, pre-existing synergies in reaching movements, and the fact that the task did not necessitate the formation of new ones. Our previous study 7 required participants to form new synergies among joint angles in the arm to intercept a ball. Greater variability in joint angles was found and interpreted as more search behavior that aided the development of synergies with larger covariation in the high extrinsic group compared to the low extrinsic group. A possible hypothesis from the dynamical systems perspective would be that variability during the practice of novel skill destabilizes pre-existing synergies and increases search behavior. Variability in action performance during search originates from both intrinsic and extrinsic sources. Therefore, it is pertinent to scrutinize these sources together in order to identify their interaction on search behavior 8 . Intrinsic variability is the inherent variability in the actions of an individual over its repetitions. The effect of intrinsic variability on learning has been mostly studied as intrinsic dynamics based on prior experience in similar tasks 9 , 10 or as motor noise in a baseline learning task 11 , 12 . However, evaluation of intrinsic variability should not be based solely on the amount of variability but also on the structure of variability. When evaluating the structure of variability, studies have demonstrated positive correlation with learning for baseline end effector variability affecting task performance 13 as well as covariation in joint angles which is variability that does not affect task performance 14 . Both types of variability characterize different aspects of action performance. Their differing effects on learning across different studies might lie in the fact that in these studies the baseline task is a simple, non-challenging, reaching movement. Conversely, learning a novel skill is a challenging task that warrants adaptive processes to destabilize pre-existing synergies and form new ones. It has already been established that higher covariation of joint angles facilitates motor performance in challenging tasks 15 , 16 . Thus, we evaluated intrinsic variability as the extent of covariation in a challenging task requiring adaptive behavior and its influence on learning new synergies was studied. Motor learning entails the formation of new synergies among the degrees of freedom. Different perspectives in the motor control literature use different notions of synergies 17 , 18 . In the current paper, a dynamical systems approach was employed and synergies were defined as temporary functional organization of degrees of freedom within the joint angles of the limbs for task specific goals 19 – 21 . These synergies can be characterized by clustering and covariation of DoF within the joint space 22 . To elucidate, consider a cartesian joint space spanned by relevant joint angles involved in the task. Within this space, there is a region of all possible joint angle configurations that results in successful task completion, referred to as the Uncontrolled Manifold (UCM) 23 , 24 . Joint configurations clustering on and around this manifold represent covariation to meet the task demands and indicate formation of stable synergies. Therefore, each task has a unique UCM and needs specific synergies to achieve the respective goal. De-novo motor learning involves moving away from pre-existing synergies and searching within the joint space to find the UCM particular to the task and form a new synergy. In this motor learning perspective, search is the process of structured change in variability over repetitions during the learning process. Recent work in reinforcement learning examined search at the task level 25 , 26 . These studies demonstrate that when learning to reach an invisible diagonally elongated rectangular space with the hand, search is modulated by reward and is structured such that the variability is greater along the elongated axis than in the orthogonal dimension. Their results suggest that variability is necessary for learning and is controlled differentially depending on its effect on task performance. These studies focused on the task level (i.e., hand position), without addressing the underlying synergies or directly manipulating task variability as is done in this study. The current study involved two tasks: categorization and learning (Fig. 1 ). The categorization task was a reaching task with target jumps to induce adaptability in evaluation of intrinsic variability. The participants were divided into high and low intrinsic variability groups on the basis of their covariation in joint angles on non-jump trials. The learning task was a lateral interception task with a body machine interface wherein joint angles of the upper limb controlled a paddle on a screen to intercept a downward moving ball. The movement pattern to control the paddle in a goal-directed manner was novel, hence the task required participants to search within the joint space during practice and learn new synergies. The participants were further divided into two groups based on their practice schedule: blocked (low extrinsic) or random (high extrinsic). Thus, the study consisted of 4 groups based on variability: low intrinsic-low extrinsic (LiLe), low intrinsic-high extrinsic (LiHe), high intrinsic-low extrinsic (HiLe), and high intrinsic-high extrinsic (HiHe). This study aimed to answer four major research questions. First, do participants learn to intercept the ball and does performance in post-test and transfer vary between the groups who are differentiated on intrinsic and extrinsic variability? Second, does the amount of search during learning vary between and within the groups? Third, do new stable synergies emerge at end of practice and does the amount of covariation in DoF differ between the groups? And lastly, does the process of searching for a novel synergy differ during learning between groups, reflected by changes in the location of synergies in joint space, over learning? We hypothesized that groups with higher variability would perform poorly during practice but show better performance in post-test and transfer. This is due to the formation of new synergies with higher covariation at the end of practice following from increased search behavior during practice. An alternative hypothesis would posit that too much and too little variability is detrimental to learning; therefore the HiHe and LiLe groups would perform the poorest. Results 46 right-handed participants (mean[SD] age of 22.45[3.39] years, twenty-eight females) were recruited for the experiment from which two participants were not considered for categorization due to technical issues during measurements. Another participant was excluded for being left-handed. Further two participants were excluded after categorization during analyses of the learning task as they did not exhibit any improvement with practice. Categorization of Intrinsic Variability: The participants were categorized based on their intrinsic variability computed using the uncontrolled manifold (UCM) analysis 24 . The analysis was conducted with the left upper limb joint angles (shoulder [3 angles], elbow [2 angles], wrist [ 2 angles] and finger [2 angles]) as the elemental variables and the fingertip position as the performance variable 6 . The participants performed a standard reaching task to a target presented on a horizontal screen in front of them at a distance of 30cm. In 20% of the trials, the target would unexpectedly jump to a location approximately 10cm on the left or the right after the initiation of the movement. The jump trials were included to encourage flexibility in reach movements. The analysis was performed only on reach trials without the target jump. For all participants, VUCM was higher than VORT, confirming that the end-effector trajectory was stabilized. Covariation (VUCM), i.e., variation in joint angles not affecting task performance, was chosen as a measure of intrinsic variability in participants. Participants with high VUCM performed the task with more variable joint configurations implying greater intrinsic variability. The participants were ranked in increasing order of VUCM and classified into low and high intrinsic variability groups based on the median (Fig. 2 ). Performance in Learning Task The participants were recalled on a different day to perform the learning task once the categorization task of all participants was completed. An approximately equal number of participants from both intrinsic variability groups performed the learning task either with blocked (low extrinsic variability) or random (high extrinsic variability) practice; leading to the four groups: LiLe, HiLe, LiHe, and HiHe. Whether performance differed after practice across the four groups was evaluated by a 4-way mixed ANOVA on absolute error with test (pre, post and transfer); test condition (blocked and random) as within participant factors and intrinsic variability (low and high) and extrinsic variability (low and high) as between participant factors. The analysis revealed significant main effects of test (F(1.16,42.83) = 217.95, p < 0.001, η 2 G = 0.65) and extrinsic variability (F(1,37) = 5.63, p = 0.02, η 2 G = 0.05). Post hoc analysis on test revealed significant differences between pre-test and post-test (p < 0.001) as well as between pre-test and transfer-test (p < 0.001), see Fig. 3 . All participants had smaller errors after practicing. Furthermore, overall errors were much lower for high extrinsic variability condition than low extrinsic variability and no effect of intrinsic variability was observed. The change in performance during learning and whether it differed for the four groups was analyzed using a 3-way mixed ANOVA on absolute error with practice phase (1 to 4) as repeated measures and intrinsic variability (low and high) and extrinsic variability (blocked and random) as between group measures. Only the main effect of the phase was significant (F(2.24,85.30) = 146.14, p < 0.001, η 2 G = 0.40). All paired comparisons between the levels in phase were significant (p < 0.001) in the post hoc tests. Participants continuously improved their performance throughout the different practice phases (Fig. 3 ). Formation of Stable Synergies Whether the changes in task performance follow from the formation of new synergies was assessed using the UCM analysis. The UCM analysis partitions the variability into VUCM (covariation; compensatory variability of DoF) and VORT (performance linked variance). VUCM > VORT is generally conceived as a signature of task stabilizing synergies 23 . A 4-way mixed ANOVA was performed on the variance with variance type (VUCM and VORT); practice phase as repeated measures; extrinsic and intrinsic variability as between group measures. The logarithmic transformation of variance was used to correct for non-normal distribution 27 . Main effect of variance (F(1,38) = 29.16, p < 0.001, η 2 G = 0.03), phase (F(2.37,90.04) = 81.70, p < 0.001, η 2 G = 0.36) and extrinsic variability (F(1,38) = 9.06, p = 0.005, η 2 G = 0.14) was observed, along with an interaction effect between variance*phase (F(2.50,95.12) = 6.89, p < 0.001, η 2 G = 0.01). Overall variance was higher in high extrinsic (random practice) condition than low extrinsic (blocked practice) condition. VUCM was higher than VORT and post hoc analysis on the main effect of phase revealed p < 0.001 for all pairwise comparisons. Post hoc analysis on the interaction between variance*phase also showed significant difference between VUCM and VORT from the second phase of practice (phase 1, p = 1; phase 2, p = 0.04; phase 3, p = 0.01 and phase 4, p < 0.001) (Fig. 4 ). This implies that VORT was not lower than VUCM from the start, but VORT reduced more over practice than VUCM, indicating the emergence of synergies with practice. No effect of intrinsic variability was observed. Amount of Search within Joint Space Emergence of new synergies requires search within the joint space. The amount of inter-trial search was evaluated as the Euclidean distance within the joint space between the end joint configurations on successive trials to the same ball arrival position. To ascertain the change in search behavior during learning a 3-way mixed ANOVA on distance in joint space with practice phase as repeated measures; intrinsic and extrinsic variability as between group measures, was performed. Significant main effects of phase (F(2.27,86.20) = 55.98, p < 0.001, η 2 G = 0.25) and extrinsic variability (F(1,38) = 11.52, p = 0.002, η 2 G = 0.19) as well as an interaction effect between phase*extrinsic variability (F(2.27,86.20) = 4.39, p = 0.012, η 2 G = 0.02) were found (Fig. 5 ). Post hoc analysis on the interaction effect between phase*extrinsic variability demonstrated significantly higher search in high extrinsic (random practice) condition as compared to low extrinsic (blocked practice) only in the first phase of learning (t=-4.31; Cohen’s d = -1.43, p = 0.003). Also, no difference was found in search across phases for low extrinsic condition, however, search does significantly reduce between the initial two phases for high extrinsic condition (t = 7.10, Cohen’s d = 1.16, p < 0.001). Structure in Search: Location of Synergies The structure in the search behavior was assessed using the joint deviation vector (JDV) analysis 22 , 28 , 29 wherein projection lengths of joint configurations in individual trials are computed along the uncontrolled manifold and orthogonal to it. These joint deviation vectors were computed from a reference joint configuration for which the mean joint configuration in the last practice phase of each participant was taken. A 4-way mixed ANOVA was performed on the joint deviation vectors with dimension (JDVUCM and JDVORT) and practice phase as repeated measures and extrinsic and intrinsic variability as between group measures. Main effects of dimension (F(1,38) = 20.99, p < 0.001, η 2 G = 0.06), phase (F(1.69,64.40) = 172.32, p < 0.001, η 2 G = 0.48) and extrinsic variability (F(1,38) = 7.7, p = 0.009, η 2 G = 0.11) were revealed. Additionally, interaction effects between dimension*phase (F(1.79,67.88) = 21.93, p < 0.001, η 2 G = 0.03) and phase*extrinsic variability (F(1.69,64.40) = 4.857, p = 0.015, η 2 G = 0.03) were significant. Post hoc analysis on dimension*phase revealed no difference between JDVUCM and JDVORT in phase 1 (p = 1) but in phase 2 (p < 0.001) and phase 3 (p < 0.001) they differed significantly. Figure 6 depicts the change in both JDVUCM and JDVORT throughout practice phases. Initially, both JDV’s were comparable, but in the following practice phases, JDVORT decreased more steeply than JDVUCM. These findings suggested that participants moved towards the uncontrolled manifold expeditiously while continuing to search along the uncontrolled manifold. The post hoc analysis on phase*extrinsic variability revealed significant differences between joint deviation vectors of low and high extrinsic variability conditions in the first phase of practice (p = 0.003). There were no differences in the later practice phases (p = 1 for both phase 2 and phase 3). The results indicated that the search in joint space was in a larger region in the high extrinsic group than in the low extrinsic group in the initial phase of practice, whereas these groups did not differ in the other phases (Fig. 7 ). Discussion This study evaluated the role of different sources of variability (intrinsic to the participant and extrinsic from the practice schedule) on the search behaviour in a de-novo motor learning task wherein participants had to learn novel synergies to perform the task. All but two participants learned the task and discovered the synergies that moved the virtual paddle to intercept the falling ball. There are four key findings in this paper; first it was shown that extrinsic variability provoked search behaviour during practice but did not lead to differences in performance during test sessions. Second, no effect of intrinsic variability on search or performance was found. Third, it was demonstrated that reduction in motor variability during learning is structured; such that performance linked variability decreases faster than covariation i.e. synergies within the joint space homed in towards the UCM in the early phases of practice whereas changes along the UCM were more gradual. Lastly, formation of stable synergies in a de-novo learning task was established. Performance in a complex de-novo learning task was found to be noncontingent on both extrinsic and intrinsic variability. The novelty of this work lies in the fact that the influence of intrinsic and extrinsic variability on the search behaviour during learning was compared concurrently. Previous studies have either estimated the correlation of intrinsic variability/individual differences and learning rate 13 , 14 or the influence of practice conditions with different degrees of variability 30 , 31 . However, to the knowledge of the authors these two sources of variability had not been assessed concomitantly to delineate their contributions. In this study, effect of extrinsic variability on learning was examined by introducing blocked and random practice schedules. While intrinsic variability was evaluated as the flexibility (operationalised with VUCM) shown by participants in performing a reaching task with target jumps. It was hypothesized that in a de-novo learning task; higher intrinsic and extrinsic variability would yield increased search behaviour and higher covariation in synergies. Our results suggested high extrinsic variability groups had greater amount of search behaviour during practice and lower error across all test conditions. Yet, participants learnt to perform the task equally well irrespective of practice schedule and intrinsic variability as their performances in post-test and transfer were comparable. A key result is that no effects of intrinsic variability were found on learning outcomes and search behaviour. A detailed interpretation of intrinsic variability and how these findings fit in the literature will follow later in the discussion. Practice schedule effects on performance outcomes do not find strong support in complex discrete tasks requiring de-novo learning 32 , 33 . Our results on extrinsic variability confirm previous findings in Mehta et al. (2025) which utilised the same experimental paradigm. The effect of extrinsic variability during practice was stronger in Mehta et al. (2025), where the interaction effects between extrinsic variability and amount of search as well as variance (VUCM and VORT) were found, which were not present in this study. Nevertheless, these were small effects (η 2 G ~ 0.01), the presence of which might depend on the specifics of the participants. One of the few studies to utilise a complex skill acquisition task to show variability of practice effects was Gray et al. (2017) 34 where baseball batters trained in VR with blocked and random practice conditions. In contrast to our study, Gray et al. did not study de-novo learning since participants were skilled baseball players with substantially less variability in pre-test performance. Since our participants were not skilled in the task, pre-test performance might have shown relatively high variability which most likely confounded the differences in performance due to variability of practice. Although practice schedule affected the overall variability in joint angles during practice, it did not interact with the search behaviour. Search behaviour was studied as the structured change in variability within joint space during practice. Participants exhibited greater search along the UCM, than along the orthogonal dimension in the joint space while performing a complex de-novo learning task. Similar results in the task space have been found where search (exploration) is studied based on the task performance at the level of the end-effector position in simple reaching tasks with pre-existing synergies. These studies 25 , 26 show a reduction in the variability of end effector position with practice and a structured change such that the variability along the target dimension was higher where the task allowed it, and the variability was lower along the dimension of the target that was restricted. While reinforcement learning (RL) studies 25 , 26 on reaching tasks with elongated targets have focused on redundancy within the task space, we extend the description and analysis of search to the redundant levels of the DoF. Although the structure in variability was recognized in RL studies, changes in this structure over practice was not considered. In the present study, the change towards the UCM within the joint space was found to be predominant and occurred more rapidly in the initial phases of practice whereas the movement along the dimensions of UCM was more gradual over practice. Previous studies established structure of variability at the end-effector level in simple reaching tasks whereas the current study demonstrated the emergence of this structure through practice in a de-novo learning task requiring new synergies. The formation of novel synergies with practice is another key finding. Our results of the UCM analysis provide evidence that as participants learn to perform the task, they also stabilise new joint angle synergies. Covariation (VUCM) and variance affecting task performance (VORT) were almost equal in the first phase of practice in our task. Their difference emerged from the second phase onwards and continued to increase with practice resulting in increasingly more stable synergies over practice. These results also validate the finding from our previous study Mehta et al. (2025) 7 which utilises the same experimental paradigm. Importantly, the learning effects found in the current study showed that synergies emerged with practice, and this was also the case in our previous work. To our knowledge, these studies are the first of their kind to show the emergence of new stable synergies in de-novo learning. Existing literature has either studied formation of synergies in relatively simple tasks with limited degrees of freedom 35 or in complex tasks where existing synergies had to be adapted 36 , 37 . In these studies, covariation in joint angles was higher than variance affecting task performance from the initial trials onwards. This suggests presence of synergies at the start of the experiment, whereas we exhibit the formation of novel synergies in a de-novo learning task. No effect of intrinsic variability was found on search behavior and performance. There are conflicting results in the literature concerning the contribution of intrinsic variability to learning and performance. Ranganathan et al. (2021) 38 reported results similar to those of our study, where covariation in a baseline task did not affect task performance when a novel end-effector coordination pattern was learned in a constrained bimanual precision skill learning paradigm. Concurrently, our results are in contrast to Singh et al. (2016) 14 where baseline covariation and not task space variability in joint angles was found to be correlated to the learning rate in both kinematic and dynamic reaching adaptation tasks. On the contrary, Wu et al. (2014) 13 showed that higher task-space variability in end effector trajectory of a baseline task increased the learning rate in a reinforcement learning and force-field adaptation tasks. The contradictory results of these studies can be attributed to the differences in the learning task and operationalisation of intrinsic variability 39 . When intrinsic variability was estimated as covariation, it was found to be positively correlated to learning rate in adaptation tasks 14 but not in skill learning 38 . Further, correlation of intrinsic variability when estimated as task space variability found mixed results 13 , 40 . Another important factor influencing the results is the level of analysis – task space vs DoF space. Most studies have focused their analysis within the task space by evaluating the end-effector kinematics 13 , 40 and did not assess the underlying DoF involved 14 , 38 , which were examined in the current study. There is a need to establish a consistent definition and evaluation method for intrinsic variability to uncover its role in motor learning 39 (see also 41 , 42 for comments and guidelines to design motor learning studies). The results of this study in combination with existing literature leads to the conjecture that extrinsic variability helps in finding novel synergies 7 , 43 while intrinsic variability might influence fine tuning and adjusting of already learnt synergies 13 , 14 .We operationalised intrinsic variability as covariation in joint angles during a baseline task which implies greater variability along the UCM, while the learning task required movement to a novel UCM. Even though we aimed to use a categorisation task that forced adaptive behaviour, future studies might categorize participant groups on other measures, perhaps VORT, when examining influence of intrinsic variability on de-novo motor learning. This study had two notable limitations. First, we evaluated intrinsic variability in a planar movement within the transverse plane while the learning task required complex movements across all body planes. It is possible that our categorization task did not sufficiently challenge the participants and capture their intrinsic variability. Second, the use of inertial measurement units (IMU’s) constrained the learning task. The abundant mapping introduced in our learning task included only 4 out of the possible 7 DoFs in upper limb joint angles. The remaining joint angles were inaccurate due to interference of movement axes based on the calibration computations of the IMU signals. Additionally, a retention test could not be performed the subsequent day to evaluate learning because the calibration of the joint angle data was highly sensitive to the accurate placement of the sensors on the body. Therefore, future studies on such body machine interface paradigms could use motion capture systems better suited for accurate streaming of real time joint angle data of complex arm movements allowing incorporation of more DoF’s and measurements on multiple days as well. In conclusion, this study incorporates the influence of different sources of variability on search behaviour and the formation of novel synergies in a complex de-novo learning task. While neither extrinsic nor intrinsic variability affected task performance, extrinsic variability did induce increased search behaviour during practice. We also demonstrate that the search behaviour within joint space is structured and changes with practice resulting in the emergence of new stable synergies. Methods Sample Size The sample size was estimated in G*Power (Version 3.1) 44 using apriori computations. There was no appropriate data available to compute the power analysis on our effects. We therefore took a moderate effect size (f = 0.5) and a correlation among repeated measures of 0.4 for sample size estimation. The sample size of 44 participants was arrived at by using an alpha = 0.05 and a power of 0.95 for a repeated measures ANOVA with between factors. Ethics Statement The study was approved by the ethics committee of the University Medical Center Groningen (No R10941) and all participants provided written informed consent prior to the experiment. A nominal monetary reward of 10 euros was provided for participation. The experimental protocol was carried out in accordance with the declaration of Helsinki. Prior to the experiment the participants were provided with the details of the experiment through an information letter. General Experimental Design The participants first performed the categorization task which involved making centre-out reaching movements with their left hand while experiencing occasional target jumps. Once all the participants completed the categorization task they were divided into low and high intrinsic variability groups based on their covariation (VUCM) on the non-jump trials. Each participant was then recalled to perform the learning task with either low or high extrinsic variability (i.e., blocked or random practice condition, respectively) This task entailed learning of a mapping between the left upper limb joint angles and a virtual paddle to intercept a downward moving ball on a vertical screen in front of participants (see Fig. 1 ). Categorization Task Design The categorization task consisted of 42 reaching trials with 8 of the trials having unexpected target jumps after the initiation of the movement. The circular target (1.5 cm diameter) was displayed 30 cm away, straight across from the start position. On jump trials, the target jumps 10cm to the left or right after 200ms from initiation of the movement. The jumps were spaced such that there are at least two jumps every ten trials. The direction and trial numbers of the jumps were randomized across participants. Apparatus The apparatus used was akin to Golenia et al (2018) 16 and Wissing et. al. ( 2020 ) 29 . The targets were displayed on a large horizontal TV screen mounted on a table and a force transducer (2 cm diameter) was placed at the start position. The task was developed in Unity (Unity Technologies) and force transducer signal was read into Unity through Arduino. The force transducer was also connected to a red LED (Light Emitting Diode; visible only to the experimenter) which lit up on pressing the sensor. Two Optotrak Certus system (Northern Digital Inc., Waterloo, Canada) cameras were used to determine the position and orientation of six rigid bodies placed on the left upper limb (index finger, hand, lower arm, upper arm and shoulder) and sternum of the participants. The rigid bodies were placed on precise anatomical positions and the bony landmarks were identified 45 . The participants upper body was then gently strapped to the extended back of a chair to avoid movement of the torso during reach. The chair was placed such that the start position was at the body midline of the sagittal plane. An elbow rest was provided to fixate the arm position at the start of every trial. Procedure The participants performed reaching movements in the transverse plane by moving their index finger from the start position to the target. They were instructed to make fast and accurate movements to the target once the target colour turned white and to hold their position until the “Return to start position” text was displayed on the screen. Before the start of each trial, the participants pressed the force transducer at the start position with their index finger. Once the participant was ready (i.e. had placed their index finger on the force transducer and the LED lit up), the experimenter manually initiated data collection of Optotrak that lasted 5s. A randomized delay between 0.5s and 1.5s was introduced prior to change in colour of target, signalling the start of trial. The total time from the start of the trial to completion of the reach movement took approximately 3s. The participant held the final reach posture for the remaining time until the Optotrak data collection for the trial was completed. On completion of the data collection the participant moved back to the start position for the next trial. Data Analysis The data analyses were performed in MATLAB (MathWorks Inc. R2021b). All the Optotrak data was filtered using a second order Butterworth filter with a cut-off frequency of 5 Hz. The initiation and termination of the reaching movement was identified for each trial and the joint angles were computed 29 . Only non-jump trials were used for the uncontrolled manifold analysis. The 9 joint angles in the non-dominant upper limb were taken as the elemental variables and the 3D fingertip position was considered the performance variable. Prior to the UCM analysis, the reach movement from the trial data was identified. The start of reach was identified as the second last data point in a window of 5 data points where the velocity fell below the 5 cm/s threshold while searching backward from the timestamp of peak velocity. Similarly, the end of the trial was the subsequent data point in a window of 5 data points where the velocity fell below the 5 cm/s threshold while searching forward from the timestamp of peak velocity. This reaching movement data was then time normalised. Each participants movement variability during the reach was partitioned into variance affecting task performance (VORT) and covariation (VUCM) 6 . The computation of VUCM and VORT is mentioned in Eq. 1 and Eq. 2 respectively where J is the computed Jacobian, C is the covariance matrix, DoF is the number of degrees of freedom in the elemental variable and DV is the degrees of freedom in the performance variable. The Jacobian was computed through the regression method 46 – 48 and the null space of the Jacobian was a linear approximation of the uncontrolled manifold. Participants were then ranked in increasing order based on VUCM and the median was determined. Participants with VUCM less than the median were categorized as low variability group and the remaining participants as high variability group. \(\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:\:VUCM=\frac{trace\left(null{\left(J\right)}^{T}*C*null\right(J\left)\right)}{DOF-DV}\:\) Equation (1) \(\:VORT=\frac{trace\left(\right(J*{{\left(J\right)}^{T})}^{-1}*J*\:C*{\left(J\right)}^{T})}{DV}\:\) Eq. (2) Learning Task Design The learning task consisted of 560 interception trials presented in a single sitting. The trials were divided into 4 sessions – pre-test (16 trials), practice (512 trials), post-test (16 trials) and transfer (16 trials). The task was identical to the one performed in Mehta et al. (2025) wherein participants learnt an abundant mapping between their upper limb joint angles and a virtual paddle to successfully intercept a virtual ball (Eq. 3). The mapping was a linear combination of 4 joint angles (Shoulder Abduction-Adduction (SAA), Elbow Pronation-Supination (EPS), Elbow Flexion-Extention (EFE) and Wrist Ulnar-Radial Deviation (WUR)). The mapping was such that each joint angle contributed to +/- 4 units of paddle movement on the screen with the total distance from left to right on the interception axis being 32 units. Therefore, a combination of joint angles was required to intercept the ball at different ball arrival positions. \(\:P=\left[\begin{array}{ccccc}\:8&\:8&\:8&\:8&\:-16\end{array}\:\right]\:\:\left[\:\begin{array}{c}SAA\\\:EPS\\\:EFE\\\:WUR\\\:1\end{array}\:\right]\) Eq. (3) The task included 4 ball departure positions (BDP; Vertical Position: 10 Unity Unit; Horizontal Position +/- 7 and +/-14 Unity Units) and 4 ball arrival positions (BAP; Vertical Position: -10 Unity Unit; Horizontal Position +/- 7 and +/-14 Unity Units) with 16 possible trajectories which were either presented in a blocked schedule (BAP was blocked) or pseudorandomized across practice phases. Further, 4 different ball arrival positions (Vertical Position: -10 Unity Unit; Horizontal Position +/- 9 and +/-16 Unity Units) were introduced during the transfer test session. The trials for each BAP within the training session were divided into 4 equal phases to estimate changes across the learning process. Each phase in total consisted of 128 trials with 32 trials of each BAP. Participants from both high and low intrinsic variability groups were equally divided into blocked and random practice groups for the learning task Apparatus The task was presented on an 86" vertical screen (Riva R2, CTOUCH Europe BV) with a 60Hz refresh rate, placed at distance of 2.5m from where the participant was seated, parallel to the frontal plane. The seat did not have any armrest. The joint angles used in the mapping were computed based on the kinematic data captured at the sampling rate of 50Hz by 4 wireless inertial measurement units (MTw Awinda, Movella Technologies, Enschede, Netherlands) placed on the chest (sternum), upper arm (dorsal side between the biceps and triceps), lower arm (dorsal side close to the wrist), and the hand (dorsal side on the 3rd metacarpal) of the participant. The data was then streamed to the Unity software (Unity technologies, San Francisco, United States) to compute the mapping and control the virtual paddle. Calibration Prior to start of the learning experiment, 2 calibration procedures were performed – inertial measurement unit (IMU) calibration and range of motion (ROM) calibration. The IMU calibration was conducted in Matlab (MathWorks Inc. R2021b) to identify the correct axis of movement based on the body segment reference frames 49 , 50 . The maxima in the movement signals of respective IMUs were used to identify the joint movement axes and compute the joint angles. The calibration procedure for IMU alignment to joint segments was performed standing upright (except for wrist ulnar-radial deviation) and consisted of a static pose (neutral position) followed by three steady repetitions each of elbow flexion-extension, elbow pronation-supination, wrist flexion-extension and wrist ulnar-radial deviation. During calibration of wrist ulnar-radial deviation, the participant was seated with their lower arm and hand supported on a flat surface. This was followed by the ROM calibration to account for variations in the ROM of individuals by normalising the values for comparison of the data across participants. The procedure included the following movements – Elbow Flexion and Extension, Elbow Pronation and Supination (at 90° elbow flexion), Wrist Flexion and Extension (at 90° elbow flexion in complete pronation and 180° elbow flexion in neutral position), Wrist Ulnar Deviation and Radial Deviation (at 90° elbow flexion in complete pronation and 180° elbow flexion in neutral position), Shoulder Abduction and Adduction and Shoulder Flexion and Extension. Each movement was performed to their extreme ranges once, and the range of motion was computed within the Unity engine. Procedure After IMU calibration, the following instructions were provided to the participants: You may only move your left arm (non-dominant) specifically the shoulder, elbow and wrist to control the paddle during the experiment. Throughout the experiment lean back on the chair and sit comfortably but do not move your torso and upper body. This is a motor learning task, and you will not be able to intercept the ball in the beginning but that is acceptable. You need to learn how to control the paddle to intercept the ball and therefore you are provided with a lot of trials. There are also pauses in between the experiment to rest and you can decide when to resume after the break. Each trial will begin once the paddle is brought to the centre of the screen and do not move the paddle before the ball begins its descent. The goal of the experiment is to find the movements that allow you to intercept the ball with the paddle. If you need to make extreme movements during the experiment, try a different movement that is comfortable. Performing the task should not lead to pain in your arms. During the instructions, the neutral position to maintain the paddle at the center of the screen was also demonstrated. This start position required the joint configurations such that the elbow was in a 90° flexion and 90° supination while adjoining the hip. The wrist and shoulder were in their neutral positions. After completion of each trial, the paddle would disappear and only appear again once it was in the start position. At the start of each trial once the paddle was in the center, a delay of 700 ms to 1200 ms was introduced before the ball began to move downwards. All ball trajectories had the same vertical velocity component, and the trial duration was fixed at 2 seconds. Feedback was also provided on successful trials by changing the paddle color from white to green for 20 ms. A break was provided after every 100 trials for the participants to rest. Data Analysis The acquired time series of ball position, paddle position, and joint angles, was split into trial-wise data 7 . The data was then filtered using a second order recursive Butterworth filter with a cutoff frequency of 2.5Hz and time normalized to 100 timestamps. Trials with deviation in paddle position at the start of a trial greater than +/- 2 units from the centre were removed as outliers (1.5% of all trials across participants). All dependent measures were computed at the last timestamp of the trial i.e. once the ball arrived at the prescribed ball arrival position (BAP) or the moment it was intercepted. Task performance was evaluated based on absolute error which was computed as the absolute distance between the centre of the paddle and the centre of the ball at the end of the trial. Search was assessed in the joint space as the Euclidean distance between the joint configurations at the end of trial for two successive trials to the same ball arrival position. The uncontrolled manifold analysis 7 and joint deviation vector (JDV) analysis 22 were performed on each BAP in a particular phase separately to study the change in the structure of variability. In the UCM analysis, the four joint angles contributing to the mapping were taken as elemental variables and the paddle position was the performance variable. The computation of VUCM and VORT were done based on Eqs. 1 and 2 respectively. However, in these analyses the Jacobian was the mapping introduced in the task and did not need to be computed from the performed trials. In computation of the JDV, reference joint configuration for each BAP was computed based on the mean of trials in the last phase of practice session. The trials of the other three practice phases were projected onto the null space and orthogonal space of the reference configuration to compute JDVUCM and JDVORT, respectively. The absolute lengths of the projections were computed and summed, then normalized by the number of dimensions and the number of trials that were projected 36 . Statistical Analysis Mixed ANOVA with groups as between subject factors and practice phase or test session as repeated measures factor was performed on all dependent measures using JASP (version 0.19.0) software. Normality of the data was checked with Q-Q plots. Greenhouse Geiser corrections were performed if the assumption of sphericity was violated and Bonferroni corrections were used for post-hoc analyses. Level of significance was set at p < 0.05 and generalised eta squared values were reported as a measure of effect sizes for ANOVA’s and cohen’s d for the posthoc comparisons. Effect sizes are interpreted according to Cohen's recommendation of 0.02 for a small effect, 0.13 for a medium effect and > 0.26 for a large effect. Declarations Acknowledgements This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 956003. The authors would like to acknowledge the contributions of Mart Bekker, Izebel Van Dam, Lars De Groot, Sophie Kuitems, Karlijn Oosterveld and Merel Peppelman for assistance in data acquisition. Author Contributions A.M., J.S., D.T. and R.B. conceived the work. A.M. carried out the experiments and performed the data analyses. A.M., J.S., D.T. and R.B. contributed to writing the manuscript. All authors have reviewed the manuscript. Data and Code availability statement All the data and relevant codes used for analysis during the study will be made available at 10.5281/zenodo.15488730 on publication. Competing Interests Statement The authors declare no conflict of interest. References Dhawale, A. K., Smith, M. A. & Ölveczky, B. P. The Role of Variability in Motor Learning. Annual Review of Neuroscience 40 , 479–498 (2017). Sternad, D. It’s not (only) the mean that matters: variability, noise and exploration in skill learning. Current Opinion in Behavioral Sciences 20 , 183–195 (2018). Gray, R. The Value of Practicing the “Wrong” Way: Skill Development and Affordance Perception Through Broad Exploration of the Solution Space. (2025) doi:10.1123/jmld.2024-0062. Pacheco, M. M. & Newell, K. M. Learning a specific, individual and generalizable coordination function: evaluating the variability of practice hypothesis in motor learning. Exp Brain Res 236 , 3307–3318 (2018). Renshaw, I. et al. An ecological dynamics approach to motor learning in practice: Reframing the learning and performing relationship in high performance sport. Asian Journal of Sport and Exercise Psychology 2 , 18–26 (2022). Tuitert, I. et al. Does practicing a wide range of joint angle configurations lead to higher flexibility in a manual obstacle-avoidance target-pointing task? PLoS ONE 12 , e0181041 (2017). Mehta, A., Smith, J., Travieso, D. & Bongers, R. M. Random and Blocked Practice Schedule Affect Search for New Movement Coordination Patterns Differently. J. Motor Learn. Dev. (2025) doi:10.1123/jmld.2024-0085. Ranganathan, R. & Newell, K. M. Changing Up the Routine: Intervention-Induced Variability in Motor Learning. Exercise and Sport Sciences Reviews 41 , 64 (2013). Komar, J., Seifert, L. & Thouvarecq, R. What Variability tells us about motor expertise: measurements and perspectives from a complex system approach. Mov Sport Sci/Sci Mot 65–77 (2015) doi:10.1051/sm/2015020. Kostrubiec, V., Zanone, pier-G., Fuchs, A. & Kelso, J. A. S. Beyond the blank slate: routes to learning new coordination patterns depend on the intrinsic dynamics of the learner—experimental evidence and theoretical model. Frontiers in Human Neuroscience 6 , (2012). Thorp, E. B., Kording, K. P. & Mussa-Ivaldi, F. A. Using noise to shape motor learning. Journal of Neurophysiology 117 , 728–737 (2017). Van Beers, R. J. Motor Learning Is Optimally Tuned to the Properties of Motor Noise. Neuron 63 , 406–417 (2009). Wu, H. G., Miyamoto, Y. R., Castro, L. N. G., Ölveczky, B. P. & Smith, M. A. Temporal structure of motor variability is dynamically regulated and predicts motor learning ability. Nat Neurosci 17 , 312–321 (2014). Singh, P., Jana, S., Ghosal, A. & Murthy, A. Exploration of joint redundancy but not task space variability facilitates supervised motor learning. Proc. Natl. Acad. Sci. U.S.A. 113 , 14414–14419 (2016). de Freitas, S. M. S. F., Scholz, J. P. & Stehman, A. J. Effect of motor planning on use of motor abundance. Neuroscience Letters 417 , 66–71 (2007). Golenia, L., Schoemaker, M. M., Otten, E., Tuitert, I. & Bongers, R. M. The development of consistency and flexibility in manual pointing during middle childhood. Developmental Psychobiology 60 , 511–519 (2018). Bruton, M. & O’Dwyer, N. Synergies in coordination: a comprehensive overview of neural, computational, and behavioral approaches. Journal of Neurophysiology 120 , 2761–2774 (2018). Profeta, V. L. S. & Turvey, M. T. Bernstein’s levels of movement construction: A contemporary perspective. Human Movement Science 57 , 111–133 (2018). Kelso, J. A. S. On the coordination dynamics of (animate) moving bodies. J. Phys. Complex. 3 , 031001 (2022). Kugler, P. N., Scott Kelso, J. A. & Turvey, M. T. 1 On the Concept of Coordinative Structures as Dissipative Structures: I. Theoretical Lines of Convergence**This work was supported by NIH grants HD 01994, NS 13617 and AM 25814. in Advances in Psychology (eds. Stelmach, G. E. & Requin, J.) vol. 1 3–47 (North-Holland, 1980). Turvey, M. T. Coordination. American Psychologist 45 , 938–953 (1990). Tuitert, I., Golenia, L., Otten, E., Bootsma, R. J. & Bongers, R. M. Task constraints act at the level of synergies and at the level of end-effector kinematics in manual reaching and manual lateral interception. Journal of Experimental Psychology: Human Perception and Performance 46 , 1511–1526 (2020). Latash, M. L., Scholz, J. P. & Schöner, G. Toward a New Theory of Motor Synergies. Motor Control 11 , 276–308 (2007). Scholz, J. P. & Schöner, G. The uncontrolled manifold concept: identifying control variables for a functional task. Experimental Brain Research 126 , 289–306 (1999). Roth, A. M. et al. Reinforcement-based processes actively regulate motor exploration along redundant solution manifolds. Proc. R. Soc. B-Biol. Sci. 290 , 20231475 (2023). van Beers, R. J., Brenner, E. & Smeets, J. B. J. Random walk of motor planning in task-irrelevant dimensions. Journal of Neurophysiology 109 , 969–977 (2013). Verrel, J. Distributional properties and variance-stabilizing transformations for measures of uncontrolled manifold effects. Journal of Neuroscience Methods 191 , 166–170 (2010). Valk, T. A., Mouton, L. J., Otten, E. & Bongers, R. M. Synergies reciprocally relate end-effector and joint-angles in rhythmic pointing movements. Sci Rep 9 , 17378 (2019). Wissing, M. B. G., Golenia, L., Smith, J. & Bongers, R. M. Adjustments in end-effector trajectory and underlying joint angle synergies after a target switch: Order of adjustment is flexible. PLoS ONE 15 , 1–25 (2020). Merbah, S. & Meulemans, T. Learning a Motor Skill: Effects of Blocked Versus Random Practice a Review. Psychologica Belgica 51 , 15–48 (2011). Cardis, M., Casadio, M. & Ranganathan, R. High variability impairs motor learning regardless of whether it affects task performance. Journal of Neurophysiology 119 , 39–48 (2018). Brady, F. The Contextual Interference Effect and Sport Skills. Percept Mot Skills 106 , 461–472 (2008). Wulf, G. & Shea, C. H. Principles derived from the study of simple skills do not generalize to complex skill learning. Psychonomic Bulletin & Review 9 , 185–211 (2002). Gray, R. Transfer of Training from Virtual to Real Baseball Batting. Front Psychol 8 , 2183 (2017). Wu, Y.-H., Truglio, T. S., Zatsiorsky, V. M. & Latash, M. L. Learning to Combine High Variability With High Precision: Lack of Transfer to a Different Task. Journal of Motor Behavior 47 , 153–165 (2015). Domkin, D., Laczko, J., Djupsjöbacka, M., Jaric, S. & Latash, M. L. Joint angle variability in 3D bimanual pointing: uncontrolled manifold analysis. Exp Brain Res 163 , 44–57 (2005). Yang, J.-F., Scholz, J. P. & Latash, M. L. The role of kinematic redundancy in adaptation of reaching. Exp Brain Res 176 , 54–69 (2007). Ranganathan, R. et al. Motor Variability Prior to Learning does not Facilitate the Ability to Adopt new Movement Solutions. Neuroscience 479 , 157–168 (2021). Lopez-Fernandez, M., Sabido, R., Caballero, C. & Moreno, F. J. Relationship between initial motor variability and learning and adaptive ability. A systematic review. Neuroscience 565 , 301–311 (2025). He, K. et al. The Statistical Determinants of the Speed of Motor Learning. PLOS Computational Biology 12 , e1005023 (2016). Ranganathan, R., Tomlinson, A. D., Lokesh, R., Lin, T.-H. & Patel, P. A tale of too many tasks: task fragmentation in motor learning and a call for model task paradigms. Exp Brain Res 239 , 1–19 (2021). Ranganathan, R., Lee, M.-H. & Krishnan, C. Ten guidelines for designing motor learning studies. BJMB 16 , 112–133 (2022). Gray, R. Changes in Movement Coordination Associated With Skill Acquisition in Baseball Batting: Freezing/Freeing Degrees of Freedom and Functional Variability. Frontiers in Psychology 11 , (2020). Faul, F., Erdfelder, E., Lang, A.-G. & Buchner, A. G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods 39 , 175–191 (2007). Wu, G. et al. ISB recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion—Part II: shoulder, elbow, wrist and hand. Journal of Biomechanics 38 , 981–992 (2005). Ferreira de Freitas, S. M. S. & Peter Scholz, J. A comparison of methods for identifying the Jacobian for uncontrolled manifold variance analysis. Journal of Biomechanics 43 , 775–777 (2010). Krishnamoorthy, V., Latash, M. L., Scholz, J. P. & Zatsiorsky, V. M. Muscle synergies during shifts of the center of pressure by standing persons. Experimental Brain Research 152 , 281–292 (2003). Tuitert, I., Valk, T. A., Otten, E., Golenia, L. & Bongers, R. M. Comparing Different Methods to Create a Linear Model for Uncontrolled Manifold Analysis. Motor Control 23 , 189–204 (2019). Bonfiglio, A., Farella, E. & Bongers, R. Accuracy and Reliability of a Novel IMU-Based Functional Calibration Algorithm for Clinical 3D Wrist Joint Angle Monitoring . 6 (2024). doi:10.1109/COINS61597.2024.10622124. Bonfiglio, A., Tacconi, D., Bongers, R. M. & Farella, E. Effects of IMU sensor-to-segment calibration on clinical 3D elbow joint angles estimation. Front. Bioeng. Biotechnol. 12 , 1385750 (2024). Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 22 Oct, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 03 Jul, 2025 Reviews received at journal 01 Jul, 2025 Reviews received at journal 24 Jun, 2025 Reviewers agreed at journal 19 Jun, 2025 Reviewers agreed at journal 17 Jun, 2025 Reviewers invited by journal 15 Jun, 2025 Editor assigned by journal 14 Jun, 2025 Editor invited by journal 13 Jun, 2025 Submission checks completed at journal 12 Jun, 2025 First submitted to journal 10 Jun, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6865411","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":475979717,"identity":"ca5b5d95-2fe2-4d93-a5ff-04e9d18e1f47","order_by":0,"name":"Anadi Mehta","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAtElEQVRIiWNgGAWjYJACZiCWg7ItiNWSwGAMZUsQryWxgWgt/A3ciY8Lf9ilb7h2+PELxh1EaJE4wLvZeEZCcu6G22lmFoxniHHYAd5t0jwJzEAtOWwGjG1EaJE/wLv9N09CfboB0VoMgLYw8yQcTgBqYX5AlBbDw7ybpXnSjhvOBPqFIZEYLXLHezd+5rGplue7nfz4w8c2G8JawFEPBWwSCURoQNX9gVQdo2AUjIJRMDIAAAJVNFJYUdncAAAAAElFTkSuQmCC","orcid":"","institution":"University of Groningen, University Medical Center Groningen","correspondingAuthor":true,"prefix":"","firstName":"Anadi","middleName":"","lastName":"Mehta","suffix":""},{"id":475979718,"identity":"84a6c2ac-06d2-4aa9-8180-49009f1e8d54","order_by":1,"name":"Joanne Smith","email":"","orcid":"","institution":"University of Groningen, University Medical Center Groningen","correspondingAuthor":false,"prefix":"","firstName":"Joanne","middleName":"","lastName":"Smith","suffix":""},{"id":475979719,"identity":"207ed1a7-b7ce-48b6-bdc6-65d21980b475","order_by":2,"name":"David Travieso","email":"","orcid":"","institution":"University of Groningen, University Medical Center Groningen","correspondingAuthor":false,"prefix":"","firstName":"David","middleName":"","lastName":"Travieso","suffix":""},{"id":475979720,"identity":"1e31d4ad-68b8-4627-a444-8f3303c9ebc7","order_by":3,"name":"Raoul M. Bongers","email":"","orcid":"","institution":"University of Groningen, University Medical Center Groningen","correspondingAuthor":false,"prefix":"","firstName":"Raoul","middleName":"M.","lastName":"Bongers","suffix":""}],"badges":[],"createdAt":"2025-06-10 17:38:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6865411/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6865411/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-20895-w","type":"published","date":"2025-10-22T16:17:11+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":85757592,"identity":"1fcd8a6c-628c-483e-87dc-ac5e5f194e1a","added_by":"auto","created_at":"2025-07-01 10:56:37","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":145976,"visible":true,"origin":"","legend":"\u003cp\u003eExperimental setups; (A) Categorization Task: Horizontal center out reaching task with occasional target jumps. Solid circle represents the straight target while dashed circles represent the target location after jump. (B) Learning Task: Lateral interception task with wireless inertial measurement units (IMU’s). The ball (green, solid circle) and the paddle (red, solid rectangle) along with the possible ball departure positions (dashed circles on top) and ball arrival positions (dashed circles on the bottom). Dashed arrow and solid arrow represent movement of the ball and paddle respectively.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/d680d97693ebdae40103daa1.png"},{"id":85757586,"identity":"eed3f830-8c39-4b08-a1fb-3af7932fd0b1","added_by":"auto","created_at":"2025-07-01 10:56:37","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":45615,"visible":true,"origin":"","legend":"\u003cp\u003eCategorization based on intrinsic variability. Participants were classified into low (blue, left of median) and high (red, right of median) intrinsic variability groups based on the median (black dashed line).\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/a959ad05670420f1546daebd.png"},{"id":85758214,"identity":"7c610fd9-70ef-498a-a659-81b9247b1b90","added_by":"auto","created_at":"2025-07-01 11:04:37","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":110241,"visible":true,"origin":"","legend":"\u003cp\u003eTask performance across sessions for all the groups - low intrinsic low extrinsic (LiLe), high intrinsic low extrinsic (HiLe), low intrinsic high extrinsic (LiHe) and high intrinsic high extrinsic (HiHe). Open circles represent averages over participants; the error bars represent the standard deviation and data points represent individual participants. Black dashed line is the threshold for success. ***p\u0026lt;0.001.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/e79b2ee4c2f563f3dd8c4750.png"},{"id":85756297,"identity":"8e26e907-a900-4edf-89cd-8938911be168","added_by":"auto","created_at":"2025-07-01 10:48:37","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":88813,"visible":true,"origin":"","legend":"\u003cp\u003eChange in mean joint angle variances across practice phases. Open circles represent averages over participants; the error bars represent the standard deviation and data points represent individual participants. VUCM: Covariation; VORT: Variance affecting task performance. *p\u0026lt;0.05 and ***p\u0026lt;0.001\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/a6e48b148c20d19e9d9356e5.png"},{"id":85757587,"identity":"60b676bb-1b78-4747-9bcb-92573f83c8ed","added_by":"auto","created_at":"2025-07-01 10:56:37","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":99047,"visible":true,"origin":"","legend":"\u003cp\u003eEuclidean distance within the joint space between the end joint configurations on successive trials to the same ball arrival position across practice phases for all groups - low intrinsic low extrinsic (LiLe), high intrinsic low extrinsic (HiLe), low intrinsic high extrinsic (LiHe) and high intrinsic high extrinsic (HiHe). Open circles represent averages over participants; the error bars represent the standard deviation and data points represent individual participants. **p\u0026lt;0.01 and ***p\u0026lt;0.001.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/1811dbda7a808695f04e331a.png"},{"id":85757591,"identity":"6c2ca944-30bb-43ff-956f-62c30b5f562d","added_by":"auto","created_at":"2025-07-01 10:56:37","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":52464,"visible":true,"origin":"","legend":"\u003cp\u003eChange across practice phases of mean joint deviation vectors in both dimensions (UCM and ORT). Open circles represent averages over participants; data points represent individual participants; and the error bars represent the standard deviation. ***p\u0026lt;0.001.\u003c/p\u003e","description":"","filename":"floatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/2e95bcd8df1b0a282ca340c0.png"},{"id":85757589,"identity":"28f9afef-aa78-430a-8c57-c32bffbbf603","added_by":"auto","created_at":"2025-07-01 10:56:37","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":73335,"visible":true,"origin":"","legend":"\u003cp\u003eChange across practice phases of mean joint deviation vectors for low and high extrinsic variability groups. Open circles represent averages over participants; data points represent individual participants; and the error bars represent the standard deviation. **p\u0026lt;0.01.\u003c/p\u003e","description":"","filename":"floatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/303b85470aa1807cf554c691.png"},{"id":94490667,"identity":"e54734ae-c1e6-4cc7-bc8d-568621341031","added_by":"auto","created_at":"2025-10-27 17:13:36","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1063561,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6865411/v1/8e36aee4-81fb-46a5-b68f-2c32528cc47b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Extrinsic Variability but not Intrinsic Variability Affects the Search for Synergies in De-novo Motor Learning","fulltext":[{"header":"Introduction","content":"\u003cp\u003eHumans have the unique ability to increase their motor repertoire by learning complex movements in which the degrees of freedom in the neuromotor system are coordinated in novel patterns to perform new tasks. The abundant nature of the human movement system which provides more degrees of freedom (DoF) than are necessary for executing a certain action, is instrumental in learning these novel coordination patterns. For instance, one can intercept a ball at a certain location in 3D space through a multitude of movements by moving one\u0026rsquo;s arm along 7 different axes of rotations (DoF at the joint angle level of analysis) across 3 different joints (shoulder, elbow and wrist). These DoFs are believed to be organized as synergies that facilitate variability without compromising task execution. This movement variability can be regulated for learning new motor skills requiring novel synergies \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Our study is an attempt to identify how different sources of variability influence the search for new stable synergies when learning a de-novo task using the dynamical systems approach.\u003c/p\u003e \u003cp\u003eJoint angle variability in performing actions can either be externally imposed through task manipulations (extrinsic) or can originate within a person (intrinsic). Extrinsic variability in movements is primarily studied by introducing differing practice conditions. Within the dynamical systems approach, variability of practice is considered to provide opportunities for exploration related to searching new movement solutions and stimulating more adaptive behavior \u003csup\u003e\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. A randomized practice schedule (high extrinsic variability) necessitates varied movements due to changing task conditions as compared to blocked practice (low extrinsic variability). Tuitert et al.\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e performed a reaching study with variable heights of obstacles and reported an increased range of motion in the joint angles but no change in covariation (i.e., compensatory variation) of the DoF within synergies during practice. This result may be attributed to the presence of stable, pre-existing synergies in reaching movements, and the fact that the task did not necessitate the formation of new ones. Our previous study \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e required participants to form new synergies among joint angles in the arm to intercept a ball. Greater variability in joint angles was found and interpreted as more search behavior that aided the development of synergies with larger covariation in the high extrinsic group compared to the low extrinsic group. A possible hypothesis from the dynamical systems perspective would be that variability during the practice of novel skill destabilizes pre-existing synergies and increases search behavior. Variability in action performance during search originates from both intrinsic and extrinsic sources. Therefore, it is pertinent to scrutinize these sources together in order to identify their interaction on search behavior \u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eIntrinsic variability is the inherent variability in the actions of an individual over its repetitions. The effect of intrinsic variability on learning has been mostly studied as intrinsic dynamics based on prior experience in similar tasks \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e or as motor noise in a baseline learning task \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. However, evaluation of intrinsic variability should not be based solely on the amount of variability but also on the structure of variability. When evaluating the structure of variability, studies have demonstrated positive correlation with learning for baseline end effector variability affecting task performance \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e as well as covariation in joint angles which is variability that does not affect task performance \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. Both types of variability characterize different aspects of action performance. Their differing effects on learning across different studies might lie in the fact that in these studies the baseline task is a simple, non-challenging, reaching movement. Conversely, learning a novel skill is a challenging task that warrants adaptive processes to destabilize pre-existing synergies and form new ones. It has already been established that higher covariation of joint angles facilitates motor performance in challenging tasks \u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e,\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Thus, we evaluated intrinsic variability as the extent of covariation in a challenging task requiring adaptive behavior and its influence on learning new synergies was studied.\u003c/p\u003e \u003cp\u003eMotor learning entails the formation of new synergies among the degrees of freedom. Different perspectives in the motor control literature use different notions of synergies \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e,\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. In the current paper, a dynamical systems approach was employed and synergies were defined as temporary functional organization of degrees of freedom within the joint angles of the limbs for task specific goals \u003csup\u003e\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. These synergies can be characterized by clustering and covariation of DoF within the joint space \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. To elucidate, consider a cartesian joint space spanned by relevant joint angles involved in the task. Within this space, there is a region of all possible joint angle configurations that results in successful task completion, referred to as the Uncontrolled Manifold (UCM) \u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e,\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. Joint configurations clustering on and around this manifold represent covariation to meet the task demands and indicate formation of stable synergies. Therefore, each task has a unique UCM and needs specific synergies to achieve the respective goal. De-novo motor learning involves moving away from pre-existing synergies and searching within the joint space to find the UCM particular to the task and form a new synergy.\u003c/p\u003e \u003cp\u003eIn this motor learning perspective, search is the process of structured change in variability over repetitions during the learning process. Recent work in reinforcement learning examined search at the task level \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e. These studies demonstrate that when learning to reach an invisible diagonally elongated rectangular space with the hand, search is modulated by reward and is structured such that the variability is greater along the elongated axis than in the orthogonal dimension. Their results suggest that variability is necessary for learning and is controlled differentially depending on its effect on task performance. These studies focused on the task level (i.e., hand position), without addressing the underlying synergies or directly manipulating task variability as is done in this study.\u003c/p\u003e \u003cp\u003eThe current study involved two tasks: categorization and learning (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). The categorization task was a reaching task with target jumps to induce adaptability in evaluation of intrinsic variability. The participants were divided into high and low intrinsic variability groups on the basis of their covariation in joint angles on non-jump trials. The learning task was a lateral interception task with a body machine interface wherein joint angles of the upper limb controlled a paddle on a screen to intercept a downward moving ball. The movement pattern to control the paddle in a goal-directed manner was novel, hence the task required participants to search within the joint space during practice and learn new synergies. The participants were further divided into two groups based on their practice schedule: blocked (low extrinsic) or random (high extrinsic). Thus, the study consisted of 4 groups based on variability: low intrinsic-low extrinsic (LiLe), low intrinsic-high extrinsic (LiHe), high intrinsic-low extrinsic (HiLe), and high intrinsic-high extrinsic (HiHe).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThis study aimed to answer four major research questions. First, do participants learn to intercept the ball and does performance in post-test and transfer vary between the groups who are differentiated on intrinsic and extrinsic variability? Second, does the amount of search during learning vary between and within the groups? Third, do new stable synergies emerge at end of practice and does the amount of covariation in DoF differ between the groups? And lastly, does the process of searching for a novel synergy differ during learning between groups, reflected by changes in the location of synergies in joint space, over learning? We hypothesized that groups with higher variability would perform poorly during practice but show better performance in post-test and transfer. This is due to the formation of new synergies with higher covariation at the end of practice following from increased search behavior during practice. An alternative hypothesis would posit that too much and too little variability is detrimental to learning; therefore the HiHe and LiLe groups would perform the poorest.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e46 right-handed participants (mean[SD] age of 22.45[3.39] years, twenty-eight females) were recruited for the experiment from which two participants were not considered for categorization due to technical issues during measurements. Another participant was excluded for being left-handed. Further two participants were excluded after categorization during analyses of the learning task as they did not exhibit any improvement with practice.\u003c/p\u003e \u003cp\u003eCategorization of Intrinsic Variability:\u003c/p\u003e \u003cp\u003eThe participants were categorized based on their intrinsic variability computed using the uncontrolled manifold (UCM) analysis \u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. The analysis was conducted with the left upper limb joint angles (shoulder [3 angles], elbow [2 angles], wrist [ 2 angles] and finger [2 angles]) as the elemental variables and the fingertip position as the performance variable \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. The participants performed a standard reaching task to a target presented on a horizontal screen in front of them at a distance of 30cm. In 20% of the trials, the target would unexpectedly jump to a location approximately 10cm on the left or the right after the initiation of the movement. The jump trials were included to encourage flexibility in reach movements. The analysis was performed only on reach trials without the target jump. For all participants, VUCM was higher than VORT, confirming that the end-effector trajectory was stabilized. Covariation (VUCM), i.e., variation in joint angles not affecting task performance, was chosen as a measure of intrinsic variability in participants. Participants with high VUCM performed the task with more variable joint configurations implying greater intrinsic variability. The participants were ranked in increasing order of VUCM and classified into low and high intrinsic variability groups based on the median (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003ePerformance in Learning Task\u003c/p\u003e \u003cp\u003eThe participants were recalled on a different day to perform the learning task once the categorization task of all participants was completed. An approximately equal number of participants from both intrinsic variability groups performed the learning task either with blocked (low extrinsic variability) or random (high extrinsic variability) practice; leading to the four groups: LiLe, HiLe, LiHe, and HiHe.\u003c/p\u003e \u003cp\u003eWhether performance differed after practice across the four groups was evaluated by a 4-way mixed ANOVA on absolute error with test (pre, post and transfer); test condition (blocked and random) as within participant factors and intrinsic variability (low and high) and extrinsic variability (low and high) as between participant factors. The analysis revealed significant main effects of test (F(1.16,42.83)\u0026thinsp;=\u0026thinsp;217.95, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.65) and extrinsic variability (F(1,37)\u0026thinsp;=\u0026thinsp;5.63, p\u0026thinsp;=\u0026thinsp;0.02, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.05). Post hoc analysis on test revealed significant differences between pre-test and post-test (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) as well as between pre-test and transfer-test (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001), see Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. All participants had smaller errors after practicing. Furthermore, overall errors were much lower for high extrinsic variability condition than low extrinsic variability and no effect of intrinsic variability was observed.\u003c/p\u003e \u003cp\u003eThe change in performance during learning and whether it differed for the four groups was analyzed using a 3-way mixed ANOVA on absolute error with practice phase (1 to 4) as repeated measures and intrinsic variability (low and high) and extrinsic variability (blocked and random) as between group measures. Only the main effect of the phase was significant (F(2.24,85.30)\u0026thinsp;=\u0026thinsp;146.14, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.40). All paired comparisons between the levels in phase were significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) in the post hoc tests. Participants continuously improved their performance throughout the different practice phases (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFormation of Stable Synergies\u003c/p\u003e \u003cp\u003eWhether the changes in task performance follow from the formation of new synergies was assessed using the UCM analysis. The UCM analysis partitions the variability into VUCM (covariation; compensatory variability of DoF) and VORT (performance linked variance). VUCM\u0026thinsp;\u0026gt;\u0026thinsp;VORT is generally conceived as a signature of task stabilizing synergies \u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. A 4-way mixed ANOVA was performed on the variance with variance type (VUCM and VORT); practice phase as repeated measures; extrinsic and intrinsic variability as between group measures. The logarithmic transformation of variance was used to correct for non-normal distribution \u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e. Main effect of variance (F(1,38)\u0026thinsp;=\u0026thinsp;29.16, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.03), phase (F(2.37,90.04)\u0026thinsp;=\u0026thinsp;81.70, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.36) and extrinsic variability (F(1,38)\u0026thinsp;=\u0026thinsp;9.06, p\u0026thinsp;=\u0026thinsp;0.005, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.14) was observed, along with an interaction effect between variance*phase (F(2.50,95.12)\u0026thinsp;=\u0026thinsp;6.89, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.01). Overall variance was higher in high extrinsic (random practice) condition than low extrinsic (blocked practice) condition. VUCM was higher than VORT and post hoc analysis on the main effect of phase revealed p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 for all pairwise comparisons. Post hoc analysis on the interaction between variance*phase also showed significant difference between VUCM and VORT from the second phase of practice (phase 1, p\u0026thinsp;=\u0026thinsp;1; phase 2, p\u0026thinsp;=\u0026thinsp;0.04; phase 3, p\u0026thinsp;=\u0026thinsp;0.01 and phase 4, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). This implies that VORT was not lower than VUCM from the start, but VORT reduced more over practice than VUCM, indicating the emergence of synergies with practice. No effect of intrinsic variability was observed.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAmount of Search within Joint Space\u003c/p\u003e \u003cp\u003eEmergence of new synergies requires search within the joint space. The amount of inter-trial search was evaluated as the Euclidean distance within the joint space between the end joint configurations on successive trials to the same ball arrival position. To ascertain the change in search behavior during learning a 3-way mixed ANOVA on distance in joint space with practice phase as repeated measures; intrinsic and extrinsic variability as between group measures, was performed. Significant main effects of phase (F(2.27,86.20)\u0026thinsp;=\u0026thinsp;55.98, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.25) and extrinsic variability (F(1,38)\u0026thinsp;=\u0026thinsp;11.52, p\u0026thinsp;=\u0026thinsp;0.002, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.19) as well as an interaction effect between phase*extrinsic variability (F(2.27,86.20)\u0026thinsp;=\u0026thinsp;4.39, p\u0026thinsp;=\u0026thinsp;0.012, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.02) were found (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003ePost hoc analysis on the interaction effect between phase*extrinsic variability demonstrated significantly higher search in high extrinsic (random practice) condition as compared to low extrinsic (blocked practice) only in the first phase of learning (t=-4.31; Cohen\u0026rsquo;s d = -1.43, p\u0026thinsp;=\u0026thinsp;0.003). Also, no difference was found in search across phases for low extrinsic condition, however, search does significantly reduce between the initial two phases for high extrinsic condition (t\u0026thinsp;=\u0026thinsp;7.10, Cohen\u0026rsquo;s d\u0026thinsp;=\u0026thinsp;1.16, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eStructure in Search: Location of Synergies\u003c/p\u003e \u003cp\u003eThe structure in the search behavior was assessed using the joint deviation vector (JDV) analysis \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e,\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e wherein projection lengths of joint configurations in individual trials are computed along the uncontrolled manifold and orthogonal to it. These joint deviation vectors were computed from a reference joint configuration for which the mean joint configuration in the last practice phase of each participant was taken.\u003c/p\u003e \u003cp\u003eA 4-way mixed ANOVA was performed on the joint deviation vectors with dimension (JDVUCM and JDVORT) and practice phase as repeated measures and extrinsic and intrinsic variability as between group measures. Main effects of dimension (F(1,38)\u0026thinsp;=\u0026thinsp;20.99, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.06), phase (F(1.69,64.40)\u0026thinsp;=\u0026thinsp;172.32, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.48) and extrinsic variability (F(1,38)\u0026thinsp;=\u0026thinsp;7.7, p\u0026thinsp;=\u0026thinsp;0.009, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.11) were revealed. Additionally, interaction effects between dimension*phase (F(1.79,67.88)\u0026thinsp;=\u0026thinsp;21.93, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.03) and phase*extrinsic variability (F(1.69,64.40)\u0026thinsp;=\u0026thinsp;4.857, p\u0026thinsp;=\u0026thinsp;0.015, η\u003csup\u003e2\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;0.03) were significant.\u003c/p\u003e \u003cp\u003ePost hoc analysis on dimension*phase revealed no difference between JDVUCM and JDVORT in phase 1 (p\u0026thinsp;=\u0026thinsp;1) but in phase 2 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and phase 3 (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) they differed significantly. Figure\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e depicts the change in both JDVUCM and JDVORT throughout practice phases. Initially, both JDV\u0026rsquo;s were comparable, but in the following practice phases, JDVORT decreased more steeply than JDVUCM. These findings suggested that participants moved towards the uncontrolled manifold expeditiously while continuing to search along the uncontrolled manifold.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe post hoc analysis on phase*extrinsic variability revealed significant differences between joint deviation vectors of low and high extrinsic variability conditions in the first phase of practice (p\u0026thinsp;=\u0026thinsp;0.003). There were no differences in the later practice phases (p\u0026thinsp;=\u0026thinsp;1 for both phase 2 and phase 3). The results indicated that the search in joint space was in a larger region in the high extrinsic group than in the low extrinsic group in the initial phase of practice, whereas these groups did not differ in the other phases (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study evaluated the role of different sources of variability (intrinsic to the participant and extrinsic from the practice schedule) on the search behaviour in a de-novo motor learning task wherein participants had to learn novel synergies to perform the task. All but two participants learned the task and discovered the synergies that moved the virtual paddle to intercept the falling ball. There are four key findings in this paper; first it was shown that extrinsic variability provoked search behaviour during practice but did not lead to differences in performance during test sessions. Second, no effect of intrinsic variability on search or performance was found. Third, it was demonstrated that reduction in motor variability during learning is structured; such that performance linked variability decreases faster than covariation i.e. synergies within the joint space homed in towards the UCM in the early phases of practice whereas changes along the UCM were more gradual. Lastly, formation of stable synergies in a de-novo learning task was established.\u003c/p\u003e \u003cp\u003ePerformance in a complex de-novo learning task was found to be noncontingent on both extrinsic and intrinsic variability. The novelty of this work lies in the fact that the influence of intrinsic and extrinsic variability on the search behaviour during learning was compared concurrently. Previous studies have either estimated the correlation of intrinsic variability/individual differences and learning rate \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e or the influence of practice conditions with different degrees of variability \u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e,\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. However, to the knowledge of the authors these two sources of variability had not been assessed concomitantly to delineate their contributions. In this study, effect of extrinsic variability on learning was examined by introducing blocked and random practice schedules. While intrinsic variability was evaluated as the flexibility (operationalised with VUCM) shown by participants in performing a reaching task with target jumps. It was hypothesized that in a de-novo learning task; higher intrinsic and extrinsic variability would yield increased search behaviour and higher covariation in synergies. Our results suggested high extrinsic variability groups had greater amount of search behaviour during practice and lower error across all test conditions. Yet, participants learnt to perform the task equally well irrespective of practice schedule and intrinsic variability as their performances in post-test and transfer were comparable. A key result is that no effects of intrinsic variability were found on learning outcomes and search behaviour. A detailed interpretation of intrinsic variability and how these findings fit in the literature will follow later in the discussion.\u003c/p\u003e \u003cp\u003ePractice schedule effects on performance outcomes do not find strong support in complex discrete tasks requiring de-novo learning \u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. Our results on extrinsic variability confirm previous findings in Mehta et al. (2025) which utilised the same experimental paradigm. The effect of extrinsic variability during practice was stronger in Mehta et al. (2025), where the interaction effects between extrinsic variability and amount of search as well as variance (VUCM and VORT) were found, which were not present in this study. Nevertheless, these were small effects (η\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e\u003csub\u003eG\u003c/sub\u003e ~ 0.01), the presence of which might depend on the specifics of the participants. One of the few studies to utilise a complex skill acquisition task to show variability of practice effects was Gray et al. (2017) \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e where baseball batters trained in VR with blocked and random practice conditions. In contrast to our study, Gray et al. did not study de-novo learning since participants were skilled baseball players with substantially less variability in pre-test performance. Since our participants were not skilled in the task, pre-test performance might have shown relatively high variability which most likely confounded the differences in performance due to variability of practice.\u003c/p\u003e \u003cp\u003eAlthough practice schedule affected the overall variability in joint angles during practice, it did not interact with the search behaviour. Search behaviour was studied as the structured change in variability within joint space during practice. Participants exhibited greater search along the UCM, than along the orthogonal dimension in the joint space while performing a complex de-novo learning task. Similar results in the task space have been found where search (exploration) is studied based on the task performance at the level of the end-effector position in simple reaching tasks with pre-existing synergies. These studies \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e show a reduction in the variability of end effector position with practice and a structured change such that the variability along the target dimension was higher where the task allowed it, and the variability was lower along the dimension of the target that was restricted. While reinforcement learning (RL) studies \u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e on reaching tasks with elongated targets have focused on redundancy within the task space, we extend the description and analysis of search to the redundant levels of the DoF. Although the structure in variability was recognized in RL studies, changes in this structure over practice was not considered. In the present study, the change towards the UCM within the joint space was found to be predominant and occurred more rapidly in the initial phases of practice whereas the movement along the dimensions of UCM was more gradual over practice. Previous studies established structure of variability at the end-effector level in simple reaching tasks whereas the current study demonstrated the emergence of this structure through practice in a de-novo learning task requiring new synergies.\u003c/p\u003e \u003cp\u003eThe formation of novel synergies with practice is another key finding. Our results of the UCM analysis provide evidence that as participants learn to perform the task, they also stabilise new joint angle synergies. Covariation (VUCM) and variance affecting task performance (VORT) were almost equal in the first phase of practice in our task. Their difference emerged from the second phase onwards and continued to increase with practice resulting in increasingly more stable synergies over practice. These results also validate the finding from our previous study Mehta et al. (2025) \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e which utilises the same experimental paradigm. Importantly, the learning effects found in the current study showed that synergies emerged with practice, and this was also the case in our previous work. To our knowledge, these studies are the first of their kind to show the emergence of new stable synergies in de-novo learning. Existing literature has either studied formation of synergies in relatively simple tasks with limited degrees of freedom \u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e or in complex tasks where existing synergies had to be adapted \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e,\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. In these studies, covariation in joint angles was higher than variance affecting task performance from the initial trials onwards. This suggests presence of synergies at the start of the experiment, whereas we exhibit the formation of novel synergies in a de-novo learning task.\u003c/p\u003e \u003cp\u003eNo effect of intrinsic variability was found on search behavior and performance. There are conflicting results in the literature concerning the contribution of intrinsic variability to learning and performance. Ranganathan et al. (2021) \u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e reported results similar to those of our study, where covariation in a baseline task did not affect task performance when a novel end-effector coordination pattern was learned in a constrained bimanual precision skill learning paradigm. Concurrently, our results are in contrast to Singh et al. (2016) \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e where baseline covariation and not task space variability in joint angles was found to be correlated to the learning rate in both kinematic and dynamic reaching adaptation tasks. On the contrary, Wu et al. (2014) \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e showed that higher task-space variability in end effector trajectory of a baseline task increased the learning rate in a reinforcement learning and force-field adaptation tasks. The contradictory results of these studies can be attributed to the differences in the learning task and operationalisation of intrinsic variability \u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. When intrinsic variability was estimated as covariation, it was found to be positively correlated to learning rate in adaptation tasks \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e but not in skill learning \u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e. Further, correlation of intrinsic variability when estimated as task space variability found mixed results \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e. Another important factor influencing the results is the level of analysis – task space vs DoF space. Most studies have focused their analysis within the task space by evaluating the end-effector kinematics \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e and did not assess the underlying DoF involved \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e, which were examined in the current study. There is a need to establish a consistent definition and evaluation method for intrinsic variability to uncover its role in motor learning\u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e (see also \u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e,\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e for comments and guidelines to design motor learning studies).\u003c/p\u003e \u003cp\u003eThe results of this study in combination with existing literature leads to the conjecture that extrinsic variability helps in finding novel synergies \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e while intrinsic variability might influence fine tuning and adjusting of already learnt synergies \u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e.We operationalised intrinsic variability as covariation in joint angles during a baseline task which implies greater variability along the UCM, while the learning task required movement to a novel UCM. Even though we aimed to use a categorisation task that forced adaptive behaviour, future studies might categorize participant groups on other measures, perhaps VORT, when examining influence of intrinsic variability on de-novo motor learning.\u003c/p\u003e \u003cp\u003eThis study had two notable limitations. First, we evaluated intrinsic variability in a planar movement within the transverse plane while the learning task required complex movements across all body planes. It is possible that our categorization task did not sufficiently challenge the participants and capture their intrinsic variability. Second, the use of inertial measurement units (IMU’s) constrained the learning task. The abundant mapping introduced in our learning task included only 4 out of the possible 7 DoFs in upper limb joint angles. The remaining joint angles were inaccurate due to interference of movement axes based on the calibration computations of the IMU signals. Additionally, a retention test could not be performed the subsequent day to evaluate learning because the calibration of the joint angle data was highly sensitive to the accurate placement of the sensors on the body. Therefore, future studies on such body machine interface paradigms could use motion capture systems better suited for accurate streaming of real time joint angle data of complex arm movements allowing incorporation of more DoF’s and measurements on multiple days as well.\u003c/p\u003e \u003cp\u003eIn conclusion, this study incorporates the influence of different sources of variability on search behaviour and the formation of novel synergies in a complex de-novo learning task. While neither extrinsic nor intrinsic variability affected task performance, extrinsic variability did induce increased search behaviour during practice. We also demonstrate that the search behaviour within joint space is structured and changes with practice resulting in the emergence of new stable synergies.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003eSample Size\u003c/p\u003e\u003cp\u003eThe sample size was estimated in G*Power (Version 3.1) \u003csup\u003e\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e using apriori computations. There was no appropriate data available to compute the power analysis on our effects. We therefore took a moderate effect size (f = 0.5) and a correlation among repeated measures of 0.4 for sample size estimation. The sample size of 44 participants was arrived at by using an alpha = 0.05 and a power of 0.95 for a repeated measures ANOVA with between factors.\u003c/p\u003e\u003cp\u003eEthics Statement\u003c/p\u003e\u003cp\u003e The study was approved by the ethics committee of the University Medical Center Groningen (No R10941) and all participants provided written informed consent prior to the experiment. A nominal monetary reward of 10 euros was provided for participation. The experimental protocol was carried out in accordance with the declaration of Helsinki. Prior to the experiment the participants were provided with the details of the experiment through an information letter.\u003c/p\u003e\u003cp\u003eGeneral Experimental Design\u003c/p\u003e\u003cp\u003eThe participants first performed the categorization task which involved making centre-out reaching movements with their left hand while experiencing occasional target jumps. Once all the participants completed the categorization task they were divided into low and high intrinsic variability groups based on their covariation (VUCM) on the non-jump trials. Each participant was then recalled to perform the learning task with either low or high extrinsic variability (i.e., blocked or random practice condition, respectively) This task entailed learning of a mapping between the left upper limb joint angles and a virtual paddle to intercept a downward moving ball on a vertical screen in front of participants (see Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eCategorization Task\u003c/p\u003e\u003ch3\u003eDesign\u003c/h3\u003e\u003cp\u003eThe categorization task consisted of 42 reaching trials with 8 of the trials having unexpected target jumps after the initiation of the movement. The circular target (1.5 cm diameter) was displayed 30 cm away, straight across from the start position. On jump trials, the target jumps 10cm to the left or right after 200ms from initiation of the movement. The jumps were spaced such that there are at least two jumps every ten trials. The direction and trial numbers of the jumps were randomized across participants.\u003c/p\u003e\u003ch3\u003eApparatus\u003c/h3\u003e\u003cp\u003eThe apparatus used was akin to Golenia et al (2018) \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e and Wissing et. al. \u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e(\u003c/span\u003e2020\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e)\u003c/span\u003e \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. The targets were displayed on a large horizontal TV screen mounted on a table and a force transducer (2 cm diameter) was placed at the start position. The task was developed in Unity (Unity Technologies) and force transducer signal was read into Unity through Arduino. The force transducer was also connected to a red LED (Light Emitting Diode; visible only to the experimenter) which lit up on pressing the sensor.\u003c/p\u003e\u003cp\u003eTwo Optotrak Certus system (Northern Digital Inc., Waterloo, Canada) cameras were used to determine the position and orientation of six rigid bodies placed on the left upper limb (index finger, hand, lower arm, upper arm and shoulder) and sternum of the participants.\u003c/p\u003e\u003cp\u003eThe rigid bodies were placed on precise anatomical positions and the bony landmarks were identified \u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e. The participants upper body was then gently strapped to the extended back of a chair to avoid movement of the torso during reach. The chair was placed such that the start position was at the body midline of the sagittal plane. An elbow rest was provided to fixate the arm position at the start of every trial.\u003c/p\u003e\u003ch3\u003eProcedure\u003c/h3\u003e\u003cp\u003eThe participants performed reaching movements in the transverse plane by moving their index finger from the start position to the target. They were instructed to make fast and accurate movements to the target once the target colour turned white and to hold their position until the “Return to start position” text was displayed on the screen. Before the start of each trial, the participants pressed the force transducer at the start position with their index finger. Once the participant was ready (i.e. had placed their index finger on the force transducer and the LED lit up), the experimenter manually initiated data collection of Optotrak that lasted 5s. A randomized delay between 0.5s and 1.5s was introduced prior to change in colour of target, signalling the start of trial. The total time from the start of the trial to completion of the reach movement took approximately 3s. The participant held the final reach posture for the remaining time until the Optotrak data collection for the trial was completed. On completion of the data collection the participant moved back to the start position for the next trial.\u003c/p\u003e\u003ch2\u003eData Analysis\u003c/h2\u003e\u003cp\u003eThe data analyses were performed in MATLAB (MathWorks Inc. R2021b). All the Optotrak data was filtered using a second order Butterworth filter with a cut-off frequency of 5 Hz. The initiation and termination of the reaching movement was identified for each trial and the joint angles were computed \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. Only non-jump trials were used for the uncontrolled manifold analysis. The 9 joint angles in the non-dominant upper limb were taken as the elemental variables and the 3D fingertip position was considered the performance variable. Prior to the UCM analysis, the reach movement from the trial data was identified. The start of reach was identified as the second last data point in a window of 5 data points where the velocity fell below the 5 cm/s threshold while searching backward from the timestamp of peak velocity. Similarly, the end of the trial was the subsequent data point in a window of 5 data points where the velocity fell below the 5 cm/s threshold while searching forward from the timestamp of peak velocity. This reaching movement data was then time normalised. Each participants movement variability during the reach was partitioned into variance affecting task performance (VORT) and covariation (VUCM) \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. The computation of VUCM and VORT is mentioned in Eq.\u0026nbsp;1 and Eq.\u0026nbsp;2 respectively where J is the computed Jacobian, C is the covariance matrix, DoF is the number of degrees of freedom in the elemental variable and DV is the degrees of freedom in the performance variable. The Jacobian was computed through the regression method \u003csup\u003e\u003cspan additionalcitationids=\"CR47\" citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e–\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e\u003c/sup\u003e and the null space of the Jacobian was a linear approximation of the uncontrolled manifold. Participants were then ranked in increasing order based on VUCM and the median was determined. Participants with VUCM less than the median were categorized as low variability group and the remaining participants as high variability group.\u003c/p\u003e\u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:\\:VUCM=\\frac{trace\\left(null{\\left(J\\right)}^{T}*C*null\\right(J\\left)\\right)}{DOF-DV}\\:\\)\u003c/span\u003e \u003c/span\u003eEquation (1)\u003c/p\u003e\u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:VORT=\\frac{trace\\left(\\right(J*{{\\left(J\\right)}^{T})}^{-1}*J*\\:C*{\\left(J\\right)}^{T})}{DV}\\:\\)\u003c/span\u003e \u003c/span\u003e Eq.\u0026nbsp;(2)\u003c/p\u003e\u003cp\u003eLearning Task\u003c/p\u003e\u003ch2\u003eDesign\u003c/h2\u003e\u003cp\u003eThe learning task consisted of 560 interception trials presented in a single sitting. The trials were divided into 4 sessions – pre-test (16 trials), practice (512 trials), post-test (16 trials) and transfer (16 trials). The task was identical to the one performed in Mehta et al. (2025) wherein participants learnt an abundant mapping between their upper limb joint angles and a virtual paddle to successfully intercept a virtual ball (Eq.\u0026nbsp;3). The mapping was a linear combination of 4 joint angles (Shoulder Abduction-Adduction (SAA), Elbow Pronation-Supination (EPS), Elbow Flexion-Extention (EFE) and Wrist Ulnar-Radial Deviation (WUR)). The mapping was such that each joint angle contributed to +/- 4 units of paddle movement on the screen with the total distance from left to right on the interception axis being 32 units. Therefore, a combination of joint angles was required to intercept the ball at different ball arrival positions.\u003c/p\u003e\u003cp\u003e \u003cspan class=\"InlineEquation\"\u003e \u003cspan class=\"mathinline\"\u003e\\(\\:P=\\left[\\begin{array}{ccccc}\\:8\u0026amp;\\:8\u0026amp;\\:8\u0026amp;\\:8\u0026amp;\\:-16\\end{array}\\:\\right]\\:\\:\\left[\\:\\begin{array}{c}SAA\\\\\\:EPS\\\\\\:EFE\\\\\\:WUR\\\\\\:1\\end{array}\\:\\right]\\)\u003c/span\u003e \u003c/span\u003e Eq.\u0026nbsp;(3)\u003c/p\u003e\u003cp\u003eThe task included 4 ball departure positions (BDP; Vertical Position: 10 Unity Unit; Horizontal Position +/- 7 and +/-14 Unity Units) and 4 ball arrival positions (BAP; Vertical Position: -10 Unity Unit; Horizontal Position +/- 7 and +/-14 Unity Units) with 16 possible trajectories which were either presented in a blocked schedule (BAP was blocked) or pseudorandomized across practice phases. Further, 4 different ball arrival positions (Vertical Position: -10 Unity Unit; Horizontal Position +/- 9 and +/-16 Unity Units) were introduced during the transfer test session. The trials for each BAP within the training session were divided into 4 equal phases to estimate changes across the learning process. Each phase in total consisted of 128 trials with 32 trials of each BAP. Participants from both high and low intrinsic variability groups were equally divided into blocked and random practice groups for the learning task\u003c/p\u003e\u003ch3\u003eApparatus\u003c/h3\u003e\u003cp\u003eThe task was presented on an 86\" vertical screen (Riva R2, CTOUCH Europe BV) with a 60Hz refresh rate, placed at distance of 2.5m from where the participant was seated, parallel to the frontal plane. The seat did not have any armrest.\u003c/p\u003e\u003cp\u003eThe joint angles used in the mapping were computed based on the kinematic data captured at the sampling rate of 50Hz by 4 wireless inertial measurement units (MTw Awinda, Movella Technologies, Enschede, Netherlands) placed on the chest (sternum), upper arm (dorsal side between the biceps and triceps), lower arm (dorsal side close to the wrist), and the hand (dorsal side on the 3rd metacarpal) of the participant. The data was then streamed to the Unity software (Unity technologies, San Francisco, United States) to compute the mapping and control the virtual paddle.\u003c/p\u003e\u003ch3\u003eCalibration\u003c/h3\u003e\u003cp\u003ePrior to start of the learning experiment, 2 calibration procedures were performed – inertial measurement unit (IMU) calibration and range of motion (ROM) calibration. The IMU calibration was conducted in Matlab (MathWorks Inc. R2021b) to identify the correct axis of movement based on the body segment reference frames \u003csup\u003e\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e,\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e\u003c/sup\u003e. The maxima in the movement signals of respective IMUs were used to identify the joint movement axes and compute the joint angles. The calibration procedure for IMU alignment to joint segments was performed standing upright (except for wrist ulnar-radial deviation) and consisted of a static pose (neutral position) followed by three steady repetitions each of elbow flexion-extension, elbow pronation-supination, wrist flexion-extension and wrist ulnar-radial deviation. During calibration of wrist ulnar-radial deviation, the participant was seated with their lower arm and hand supported on a flat surface.\u003c/p\u003e\u003cp\u003eThis was followed by the ROM calibration to account for variations in the ROM of individuals by normalising the values for comparison of the data across participants. The procedure included the following movements – Elbow Flexion and Extension, Elbow Pronation and Supination (at 90° elbow flexion), Wrist Flexion and Extension (at 90° elbow flexion in complete pronation and 180° elbow flexion in neutral position), Wrist Ulnar Deviation and Radial Deviation (at 90° elbow flexion in complete pronation and 180° elbow flexion in neutral position), Shoulder Abduction and Adduction and Shoulder Flexion and Extension. Each movement was performed to their extreme ranges once, and the range of motion was computed within the Unity engine.\u003c/p\u003e\u003cp\u003eProcedure\u003c/p\u003e\u003cp\u003eAfter IMU calibration, the following instructions were provided to the participants:\u003c/p\u003e\u003cp\u003e \u003c/p\u003e\u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eYou may only move your left arm (non-dominant) specifically the shoulder, elbow and wrist to control the paddle during the experiment. Throughout the experiment lean back on the chair and sit comfortably but do not move your torso and upper body.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThis is a motor learning task, and you will not be able to intercept the ball in the beginning but that is acceptable. You need to learn how to control the paddle to intercept the ball and therefore you are provided with a lot of trials.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThere are also pauses in between the experiment to rest and you can decide when to resume after the break.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eEach trial will begin once the paddle is brought to the centre of the screen and do not move the paddle before the ball begins its descent.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe goal of the experiment is to find the movements that allow you to intercept the ball with the paddle.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eIf you need to make extreme movements during the experiment, try a different movement that is comfortable. Performing the task should not lead to pain in your arms.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eDuring the instructions, the neutral position to maintain the paddle at the center of the screen was also demonstrated. This start position required the joint configurations such that the elbow was in a 90° flexion and 90° supination while adjoining the hip. The wrist and shoulder were in their neutral positions. After completion of each trial, the paddle would disappear and only appear again once it was in the start position. At the start of each trial once the paddle was in the center, a delay of 700 ms to 1200 ms was introduced before the ball began to move downwards. All ball trajectories had the same vertical velocity component, and the trial duration was fixed at 2 seconds. Feedback was also provided on successful trials by changing the paddle color from white to green for 20 ms. A break was provided after every 100 trials for the participants to rest.\u003c/p\u003e\u003ch2\u003eData Analysis\u003c/h2\u003e\u003cp\u003eThe acquired time series of ball position, paddle position, and joint angles, was split into trial-wise data \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. The data was then filtered using a second order recursive Butterworth filter with a cutoff frequency of 2.5Hz and time normalized to 100 timestamps. Trials with deviation in paddle position at the start of a trial greater than +/- 2 units from the centre were removed as outliers (1.5% of all trials across participants).\u003c/p\u003e\u003cp\u003eAll dependent measures were computed at the last timestamp of the trial i.e. once the ball arrived at the prescribed ball arrival position (BAP) or the moment it was intercepted.\u003c/p\u003e\u003cp\u003eTask performance was evaluated based on absolute error which was computed as the absolute distance between the centre of the paddle and the centre of the ball at the end of the trial.\u003c/p\u003e\u003cp\u003eSearch was assessed in the joint space as the Euclidean distance between the joint configurations at the end of trial for two successive trials to the same ball arrival position.\u003c/p\u003e\u003cp\u003eThe uncontrolled manifold analysis \u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e and joint deviation vector (JDV) analysis \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e were performed on each BAP in a particular phase separately to study the change in the structure of variability. In the UCM analysis, the four joint angles contributing to the mapping were taken as elemental variables and the paddle position was the performance variable. The computation of VUCM and VORT were done based on Eqs.\u0026nbsp;1 and 2 respectively. However, in these analyses the Jacobian was the mapping introduced in the task and did not need to be computed from the performed trials. In computation of the JDV, reference joint configuration for each BAP was computed based on the mean of trials in the last phase of practice session. The trials of the other three practice phases were projected onto the null space and orthogonal space of the reference configuration to compute JDVUCM and JDVORT, respectively. The absolute lengths of the projections were computed and summed, then normalized by the number of dimensions and the number of trials that were projected \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\u003ch2\u003eStatistical Analysis\u003c/h2\u003e\u003cp\u003eMixed ANOVA with groups as between subject factors and practice phase or test session as repeated measures factor was performed on all dependent measures using JASP (version 0.19.0) software.\u003c/p\u003e\u003cp\u003eNormality of the data was checked with Q-Q plots. Greenhouse Geiser corrections were performed if the assumption of sphericity was violated and Bonferroni corrections were used for post-hoc analyses. Level of significance was set at p \u0026lt; 0.05 and generalised eta squared values were reported as a measure of effect sizes for ANOVA’s and cohen’s d for the posthoc comparisons. Effect sizes are interpreted according to Cohen's recommendation of 0.02 for a small effect, 0.13 for a medium effect and \u0026gt; 0.26 for a large effect.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eThis project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 956003. The authors would like to acknowledge the contributions of Mart Bekker, Izebel Van Dam, Lars De Groot, Sophie Kuitems, Karlijn Oosterveld and Merel Peppelman for assistance in data acquisition.\u003c/p\u003e\n\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003eA.M., J.S., D.T. and R.B. conceived the work. A.M. carried out the experiments and performed the data analyses. A.M., J.S., D.T. and R.B. contributed to writing the manuscript. All authors have reviewed the manuscript.\u003c/p\u003e\n\u003ch2\u003eData and Code availability statement\u003c/h2\u003e\n\u003cp\u003eAll the data and relevant codes used for analysis during the study will be made available at 10.5281/zenodo.15488730 on publication.\u003c/p\u003e\n\u003ch2\u003eCompeting Interests Statement\u003c/h2\u003e\n\u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eDhawale, A. K., Smith, M. A. \u0026amp; \u0026Ouml;lveczky, B. P. The Role of Variability in Motor Learning. \u003cem\u003eAnnual Review of Neuroscience\u003c/em\u003e \u003cstrong\u003e40\u003c/strong\u003e, 479\u0026ndash;498 (2017).\u003c/li\u003e\n\u003cli\u003eSternad, D. It\u0026rsquo;s not (only) the mean that matters: variability, noise and exploration in skill learning. \u003cem\u003eCurrent Opinion in Behavioral Sciences\u003c/em\u003e \u003cstrong\u003e20\u003c/strong\u003e, 183\u0026ndash;195 (2018).\u003c/li\u003e\n\u003cli\u003eGray, R. The Value of Practicing the \u0026ldquo;Wrong\u0026rdquo; Way: Skill Development and Affordance Perception Through Broad Exploration of the Solution Space. (2025) doi:10.1123/jmld.2024-0062.\u003c/li\u003e\n\u003cli\u003ePacheco, M. M. \u0026amp; Newell, K. M. Learning a specific, individual and generalizable coordination function: evaluating the variability of practice hypothesis in motor learning. \u003cem\u003eExp Brain Res\u003c/em\u003e \u003cstrong\u003e236\u003c/strong\u003e, 3307\u0026ndash;3318 (2018).\u003c/li\u003e\n\u003cli\u003eRenshaw, I. \u003cem\u003eet al.\u003c/em\u003e An ecological dynamics approach to motor learning in practice: Reframing the learning and performing relationship in high performance sport. \u003cem\u003eAsian Journal of Sport and Exercise Psychology\u003c/em\u003e \u003cstrong\u003e2\u003c/strong\u003e, 18\u0026ndash;26 (2022).\u003c/li\u003e\n\u003cli\u003eTuitert, I. \u003cem\u003eet al.\u003c/em\u003e Does practicing a wide range of joint angle configurations lead to higher flexibility in a manual obstacle-avoidance target-pointing task? \u003cem\u003ePLoS ONE\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, e0181041 (2017).\u003c/li\u003e\n\u003cli\u003eMehta, A., Smith, J., Travieso, D. \u0026amp; Bongers, R. M. Random and Blocked Practice Schedule Affect Search for New Movement Coordination Patterns Differently. \u003cem\u003eJ. Motor Learn. Dev.\u003c/em\u003e (2025) doi:10.1123/jmld.2024-0085.\u003c/li\u003e\n\u003cli\u003eRanganathan, R. \u0026amp; Newell, K. M. Changing Up the Routine: Intervention-Induced Variability in Motor Learning. \u003cem\u003eExercise and Sport Sciences Reviews\u003c/em\u003e \u003cstrong\u003e41\u003c/strong\u003e, 64 (2013).\u003c/li\u003e\n\u003cli\u003eKomar, J., Seifert, L. \u0026amp; Thouvarecq, R. What Variability tells us about motor expertise: measurements and perspectives from a complex system approach. \u003cem\u003eMov Sport Sci/Sci Mot\u003c/em\u003e 65\u0026ndash;77 (2015) doi:10.1051/sm/2015020.\u003c/li\u003e\n\u003cli\u003eKostrubiec, V., Zanone, pier-G., Fuchs, A. \u0026amp; Kelso, J. A. S. Beyond the blank slate: routes to learning new coordination patterns depend on the intrinsic dynamics of the learner\u0026mdash;experimental evidence and theoretical model. \u003cem\u003eFrontiers in Human Neuroscience\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, (2012).\u003c/li\u003e\n\u003cli\u003eThorp, E. B., Kording, K. P. \u0026amp; Mussa-Ivaldi, F. A. Using noise to shape motor learning. \u003cem\u003eJournal of Neurophysiology\u003c/em\u003e \u003cstrong\u003e117\u003c/strong\u003e, 728\u0026ndash;737 (2017).\u003c/li\u003e\n\u003cli\u003eVan Beers, R. J. Motor Learning Is Optimally Tuned to the Properties of Motor Noise. \u003cem\u003eNeuron\u003c/em\u003e \u003cstrong\u003e63\u003c/strong\u003e, 406\u0026ndash;417 (2009).\u003c/li\u003e\n\u003cli\u003eWu, H. G., Miyamoto, Y. R., Castro, L. N. G., \u0026Ouml;lveczky, B. P. \u0026amp; Smith, M. A. Temporal structure of motor variability is dynamically regulated and predicts motor learning ability. \u003cem\u003eNat Neurosci\u003c/em\u003e \u003cstrong\u003e17\u003c/strong\u003e, 312\u0026ndash;321 (2014).\u003c/li\u003e\n\u003cli\u003eSingh, P., Jana, S., Ghosal, A. \u0026amp; Murthy, A. Exploration of joint redundancy but not task space variability facilitates supervised motor learning. \u003cem\u003eProc. Natl. Acad. Sci. U.S.A.\u003c/em\u003e \u003cstrong\u003e113\u003c/strong\u003e, 14414\u0026ndash;14419 (2016).\u003c/li\u003e\n\u003cli\u003ede Freitas, S. M. S. F., Scholz, J. P. \u0026amp; Stehman, A. J. Effect of motor planning on use of motor abundance. \u003cem\u003eNeuroscience Letters\u003c/em\u003e \u003cstrong\u003e417\u003c/strong\u003e, 66\u0026ndash;71 (2007).\u003c/li\u003e\n\u003cli\u003eGolenia, L., Schoemaker, M. M., Otten, E., Tuitert, I. \u0026amp; Bongers, R. M. The development of consistency and flexibility in manual pointing during middle childhood. \u003cem\u003eDevelopmental Psychobiology\u003c/em\u003e \u003cstrong\u003e60\u003c/strong\u003e, 511\u0026ndash;519 (2018).\u003c/li\u003e\n\u003cli\u003eBruton, M. \u0026amp; O\u0026rsquo;Dwyer, N. Synergies in coordination: a comprehensive overview of neural, computational, and behavioral approaches. \u003cem\u003eJournal of Neurophysiology\u003c/em\u003e \u003cstrong\u003e120\u003c/strong\u003e, 2761\u0026ndash;2774 (2018).\u003c/li\u003e\n\u003cli\u003eProfeta, V. L. S. \u0026amp; Turvey, M. T. Bernstein\u0026rsquo;s levels of movement construction: A contemporary perspective. \u003cem\u003eHuman Movement Science\u003c/em\u003e \u003cstrong\u003e57\u003c/strong\u003e, 111\u0026ndash;133 (2018).\u003c/li\u003e\n\u003cli\u003eKelso, J. A. S. On the coordination dynamics of (animate) moving bodies. \u003cem\u003eJ. Phys. Complex.\u003c/em\u003e \u003cstrong\u003e3\u003c/strong\u003e, 031001 (2022).\u003c/li\u003e\n\u003cli\u003eKugler, P. N., Scott Kelso, J. A. \u0026amp; Turvey, M. T. 1 On the Concept of Coordinative Structures as Dissipative Structures: I. Theoretical Lines of Convergence**This work was supported by NIH grants HD 01994, NS 13617 and AM 25814. in \u003cem\u003eAdvances in Psychology\u003c/em\u003e (eds. Stelmach, G. E. \u0026amp; Requin, J.) vol. 1 3\u0026ndash;47 (North-Holland, 1980).\u003c/li\u003e\n\u003cli\u003eTurvey, M. T. Coordination. \u003cem\u003eAmerican Psychologist\u003c/em\u003e \u003cstrong\u003e45\u003c/strong\u003e, 938\u0026ndash;953 (1990).\u003c/li\u003e\n\u003cli\u003eTuitert, I., Golenia, L., Otten, E., Bootsma, R. J. \u0026amp; Bongers, R. M. Task constraints act at the level of synergies and at the level of end-effector kinematics in manual reaching and manual lateral interception. \u003cem\u003eJournal of Experimental Psychology: Human Perception and Performance\u003c/em\u003e \u003cstrong\u003e46\u003c/strong\u003e, 1511\u0026ndash;1526 (2020).\u003c/li\u003e\n\u003cli\u003eLatash, M. L., Scholz, J. P. \u0026amp; Sch\u0026ouml;ner, G. Toward a New Theory of Motor Synergies. \u003cem\u003eMotor Control\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, 276\u0026ndash;308 (2007).\u003c/li\u003e\n\u003cli\u003eScholz, J. P. \u0026amp; Sch\u0026ouml;ner, G. The uncontrolled manifold concept: identifying control variables for a functional task. \u003cem\u003eExperimental Brain Research\u003c/em\u003e \u003cstrong\u003e126\u003c/strong\u003e, 289\u0026ndash;306 (1999).\u003c/li\u003e\n\u003cli\u003eRoth, A. M. \u003cem\u003eet al.\u003c/em\u003e Reinforcement-based processes actively regulate motor exploration along redundant solution manifolds. \u003cem\u003eProc. R. Soc. B-Biol. Sci.\u003c/em\u003e \u003cstrong\u003e290\u003c/strong\u003e, 20231475 (2023).\u003c/li\u003e\n\u003cli\u003evan Beers, R. J., Brenner, E. \u0026amp; Smeets, J. B. J. Random walk of motor planning in task-irrelevant dimensions. \u003cem\u003eJournal of Neurophysiology\u003c/em\u003e \u003cstrong\u003e109\u003c/strong\u003e, 969\u0026ndash;977 (2013).\u003c/li\u003e\n\u003cli\u003eVerrel, J. Distributional properties and variance-stabilizing transformations for measures of uncontrolled manifold effects. \u003cem\u003eJournal of Neuroscience Methods\u003c/em\u003e \u003cstrong\u003e191\u003c/strong\u003e, 166\u0026ndash;170 (2010).\u003c/li\u003e\n\u003cli\u003eValk, T. A., Mouton, L. J., Otten, E. \u0026amp; Bongers, R. M. Synergies reciprocally relate end-effector and joint-angles in rhythmic pointing movements. \u003cem\u003eSci Rep\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 17378 (2019).\u003c/li\u003e\n\u003cli\u003eWissing, M. B. G., Golenia, L., Smith, J. \u0026amp; Bongers, R. M. Adjustments in end-effector trajectory and underlying joint angle synergies after a target switch: Order of adjustment is flexible. \u003cem\u003ePLoS ONE\u003c/em\u003e \u003cstrong\u003e15\u003c/strong\u003e, 1\u0026ndash;25 (2020).\u003c/li\u003e\n\u003cli\u003eMerbah, S. \u0026amp; Meulemans, T. Learning a Motor Skill: Effects of Blocked Versus Random Practice a Review. \u003cem\u003ePsychologica Belgica\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, 15\u0026ndash;48 (2011).\u003c/li\u003e\n\u003cli\u003eCardis, M., Casadio, M. \u0026amp; Ranganathan, R. High variability impairs motor learning regardless of whether it affects task performance. \u003cem\u003eJournal of Neurophysiology\u003c/em\u003e \u003cstrong\u003e119\u003c/strong\u003e, 39\u0026ndash;48 (2018).\u003c/li\u003e\n\u003cli\u003eBrady, F. The Contextual Interference Effect and Sport Skills. \u003cem\u003ePercept Mot Skills\u003c/em\u003e \u003cstrong\u003e106\u003c/strong\u003e, 461\u0026ndash;472 (2008).\u003c/li\u003e\n\u003cli\u003eWulf, G. \u0026amp; Shea, C. H. Principles derived from the study of simple skills do not generalize to complex skill learning. \u003cem\u003ePsychonomic Bulletin \u0026amp; Review\u003c/em\u003e \u003cstrong\u003e9\u003c/strong\u003e, 185\u0026ndash;211 (2002).\u003c/li\u003e\n\u003cli\u003eGray, R. Transfer of Training from Virtual to Real Baseball Batting. \u003cem\u003eFront Psychol\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, 2183 (2017).\u003c/li\u003e\n\u003cli\u003eWu, Y.-H., Truglio, T. S., Zatsiorsky, V. M. \u0026amp; Latash, M. L. Learning to Combine High Variability With High Precision: Lack of Transfer to a Different Task. \u003cem\u003eJournal of Motor Behavior\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, 153\u0026ndash;165 (2015).\u003c/li\u003e\n\u003cli\u003eDomkin, D., Laczko, J., Djupsj\u0026ouml;backa, M., Jaric, S. \u0026amp; Latash, M. L. Joint angle variability in 3D bimanual pointing: uncontrolled manifold analysis. \u003cem\u003eExp Brain Res\u003c/em\u003e \u003cstrong\u003e163\u003c/strong\u003e, 44\u0026ndash;57 (2005).\u003c/li\u003e\n\u003cli\u003eYang, J.-F., Scholz, J. P. \u0026amp; Latash, M. L. The role of kinematic redundancy in adaptation of reaching. \u003cem\u003eExp Brain Res\u003c/em\u003e \u003cstrong\u003e176\u003c/strong\u003e, 54\u0026ndash;69 (2007).\u003c/li\u003e\n\u003cli\u003eRanganathan, R. \u003cem\u003eet al.\u003c/em\u003e Motor Variability Prior to Learning does not Facilitate the Ability to Adopt new Movement Solutions. \u003cem\u003eNeuroscience\u003c/em\u003e \u003cstrong\u003e479\u003c/strong\u003e, 157\u0026ndash;168 (2021).\u003c/li\u003e\n\u003cli\u003eLopez-Fernandez, M., Sabido, R., Caballero, C. \u0026amp; Moreno, F. J. Relationship between initial motor variability and learning and adaptive ability. A systematic review. \u003cem\u003eNeuroscience\u003c/em\u003e \u003cstrong\u003e565\u003c/strong\u003e, 301\u0026ndash;311 (2025).\u003c/li\u003e\n\u003cli\u003eHe, K. \u003cem\u003eet al.\u003c/em\u003e The Statistical Determinants of the Speed of Motor Learning. \u003cem\u003ePLOS Computational Biology\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, e1005023 (2016).\u003c/li\u003e\n\u003cli\u003eRanganathan, R., Tomlinson, A. D., Lokesh, R., Lin, T.-H. \u0026amp; Patel, P. A tale of too many tasks: task fragmentation in motor learning and a call for model task paradigms. \u003cem\u003eExp Brain Res\u003c/em\u003e \u003cstrong\u003e239\u003c/strong\u003e, 1\u0026ndash;19 (2021).\u003c/li\u003e\n\u003cli\u003eRanganathan, R., Lee, M.-H. \u0026amp; Krishnan, C. Ten guidelines for designing motor learning studies. \u003cem\u003eBJMB\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 112\u0026ndash;133 (2022).\u003c/li\u003e\n\u003cli\u003eGray, R. Changes in Movement Coordination Associated With Skill Acquisition in Baseball Batting: Freezing/Freeing Degrees of Freedom and Functional Variability. \u003cem\u003eFrontiers in Psychology\u003c/em\u003e \u003cstrong\u003e11\u003c/strong\u003e, (2020).\u003c/li\u003e\n\u003cli\u003eFaul, F., Erdfelder, E., Lang, A.-G. \u0026amp; Buchner, A. G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. \u003cem\u003eBehavior Research Methods\u003c/em\u003e \u003cstrong\u003e39\u003c/strong\u003e, 175\u0026ndash;191 (2007).\u003c/li\u003e\n\u003cli\u003eWu, G. \u003cem\u003eet al.\u003c/em\u003e ISB recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion\u0026mdash;Part II: shoulder, elbow, wrist and hand. \u003cem\u003eJournal of Biomechanics\u003c/em\u003e \u003cstrong\u003e38\u003c/strong\u003e, 981\u0026ndash;992 (2005).\u003c/li\u003e\n\u003cli\u003eFerreira de Freitas, S. M. S. \u0026amp; Peter Scholz, J. A comparison of methods for identifying the Jacobian for uncontrolled manifold variance analysis. \u003cem\u003eJournal of Biomechanics\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 775\u0026ndash;777 (2010).\u003c/li\u003e\n\u003cli\u003eKrishnamoorthy, V., Latash, M. L., Scholz, J. P. \u0026amp; Zatsiorsky, V. M. Muscle synergies during shifts of the center of pressure by standing persons. \u003cem\u003eExperimental Brain Research\u003c/em\u003e \u003cstrong\u003e152\u003c/strong\u003e, 281\u0026ndash;292 (2003).\u003c/li\u003e\n\u003cli\u003eTuitert, I., Valk, T. A., Otten, E., Golenia, L. \u0026amp; Bongers, R. M. Comparing Different Methods to Create a Linear Model for Uncontrolled Manifold Analysis. \u003cem\u003eMotor Control\u003c/em\u003e \u003cstrong\u003e23\u003c/strong\u003e, 189\u0026ndash;204 (2019).\u003c/li\u003e\n\u003cli\u003eBonfiglio, A., Farella, E. \u0026amp; Bongers, R. \u003cem\u003eAccuracy and Reliability of a Novel IMU-Based Functional Calibration Algorithm for Clinical 3D Wrist Joint Angle Monitoring\u003c/em\u003e. 6 (2024). doi:10.1109/COINS61597.2024.10622124.\u003c/li\u003e\n\u003cli\u003eBonfiglio, A., Tacconi, D., Bongers, R. M. \u0026amp; Farella, E. Effects of IMU sensor-to-segment calibration on clinical 3D elbow joint angles estimation. \u003cem\u003eFront. Bioeng. Biotechnol.\u003c/em\u003e\u003cstrong\u003e12\u003c/strong\u003e, 1385750 (2024).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Motor Learning, Synergies, Exploration, Intrinsic Variability, Practice Variability and Uncontrolled Manifold Analysis","lastPublishedDoi":"10.21203/rs.3.rs-6865411/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6865411/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDe-novo motor learning involves emergence of novel task specific joint angle coordination patterns, called synergies, through structured change in joint angle variability, called search. Search implies variability in joint angles and task performance, but the understanding of the relationship between variability and learning is still unclear. This study examines the contributions of extrinsic variability - differing practice schedules in the task (blocked vs random) - and intrinsic variability - flexibility in individual\u0026rsquo;s movements \u0026ndash; on learning, search and synergy formation. Participants were first categorized based on their intrinsic variability (i.e., joint angle covariation in an adaptive reaching task). Learning was then evaluated on a virtual lateral interception task with a redundant mapping between joint angles and virtual paddle position that necessitated forming new synergies. Overall, search reduced, and synergies emerged with practice. Intrinsic variability did not affect learning, search or synergy formation. Extrinsic variability affected search during practice, but did not interact with emergence of synergies. During practice, the uncontrolled manifold (solution sub-space within the joint space) was approached rapidly in the initial practice phases while the movement along the uncontrolled manifold was more gradual. We demonstrate search in joint space during learning is structured and extrinsic variability increases search behaviour.\u003c/p\u003e","manuscriptTitle":"Extrinsic Variability but not Intrinsic Variability Affects the Search for Synergies in De-novo Motor Learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-01 10:48:32","doi":"10.21203/rs.3.rs-6865411/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-07-03T12:50:18+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-01T14:28:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-24T23:07:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"323364006608909119700963091964218949317","date":"2025-06-19T16:25:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"127790531972751878591098382886034519898","date":"2025-06-17T09:38:35+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-06-15T08:57:38+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-06-14T05:26:21+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-06-13T21:21:44+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-06-12T08:18:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-06-10T17:22:28+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"be11e0cd-560a-499f-b7ab-60f8e971a752","owner":[],"postedDate":"July 1st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":50543395,"name":"Biological sciences/Psychology"},{"id":50543396,"name":"Biological sciences/Psychology/Human behaviour"},{"id":50543397,"name":"Biological sciences/Neuroscience/Motor control"}],"tags":[],"updatedAt":"2025-10-27T16:28:16+00:00","versionOfRecord":{"articleIdentity":"rs-6865411","link":"https://doi.org/10.1038/s41598-025-20895-w","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-10-22 16:17:11","publishedOnDateReadable":"October 22nd, 2025"},"versionCreatedAt":"2025-07-01 10:48:32","video":"","vorDoi":"10.1038/s41598-025-20895-w","vorDoiUrl":"https://doi.org/10.1038/s41598-025-20895-w","workflowStages":[]},"version":"v1","identity":"rs-6865411","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6865411","identity":"rs-6865411","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}