Anticipatory control of digit kinematics: a developmental milestone for motor skill acquisition

Abstract

During development, children naturally explore and manipulate objects with their hands and fingers, becoming more skillful with practice and with age. Adults grip objects strategically and adjust forces based on task demands; for example, digit positions on a glass of milk differ when transporting it versus when drinking from it. In these interactions, sensory feedback about digit position is crucial for precise control of the object. The brain forms distinct sensorimotor memories for both digit forces and positions, utilizing them to finely adjust grip parameters during skilled object manipulation. While the development of digit forces is well-studied, the development of anticipatory control of digit position remains less understood. This study investigated the development of anticipatory control of digit positions in children 5–7, 8–10, and 15–17 years during a dexterous manipulation task. A modified grip apparatus measured digit forces, positions, and object tilt (i.e., performance) at lift-off. A total of 39 children completed the study. Children aged 5–7 years failed to use sensorimotor memories about the object’s hidden mass location from previous trials to plan their digit positions and digit forces. They applied an inaccurate torque that resulted in significant object tilt. Children aged 8–10 years showed a modest ability to use sensorimotor memories from previous trials when compared with the adolescent group. This study elucidates the ongoing development of dexterity into adolescence, offering insight into the maturation of sensorimotor capabilities throughout childhood and adolescence.

Introduction

Development of fine motor skills is essential to gain independence in manual activities of daily living as well as for social and play skills associated with cognitive development, e.g., feeding, bathing, grooming, drawing, writing, typing (Fischer et al. 2020; Palmis et al. 2017; van der Fels et al. 2015; Winter et al. 2021). Skillful object manipulation requires a complex interplay between the magnitude of forces exerted on an object (i.e., digit forces) and where these forces are exerted (i.e., digit positioning on the object; Fu et al. 2010; Lukos et al. 2007; Parikh et al. 2020; Zhang et al. 2010). Such a manipulation depends on anticipatory mechanisms based on sensorimotor memories of the object properties and/or task, which are formed through experiences (Gordon et al. 1993; Johansson and Cole 1992). To date, the literature characterizing the development of skillful object manipulation has focused on studying the scaling of digit forces to object properties and task conditions (Dayanidhi et al. 2013; Forssberg et al. 1991, 1992; Gordon et al. 1993; Gordon 2001; Johansson and Westling 1988). The presence of anticipatory control was determined based on the scaling of grip (horizontal) and load (vertical) forces just before an object is lifted (Gordon et al. 1993; Johansson and Cole 1992; Parikh et al. 2020).

Through a series of experiments, researchers demonstrated that at about two years of age, typically developing children can use a precision grip to lift objects, exhibiting grip and load force profiles that indicate the emergence of anticipatory control (Forssberg et al., 1991; Forssberg et al., 1992). By approximately 6–8 years of age, their force coordination strategy more closely matches that observed in typically developed adults (Forssberg et al., 1992). However, in a child’s typical environment, they are free to grasp and manipulate objects as they choose and may choose a variety of digit positioning strategies not studied in previous force coordination experiments. In other experiments, children 5–10 years old increasingly choose grip strategies that maximize their “end state comfort,” depending on the desired object manipulation (Wunsch and Weigelt 2016). This adaptive choice suggests that the development of anticipatory mechanisms for dexterous manipulation in children is likely related to their ability to modulate both digit positions and digit forces.

The coordination between digit forces and digit position is a ubiquitous strategy and fundamental feature of grasp control in our daily manipulations of objects (Davare et al. 2019; Fu et al. 2010; Lee-Miller et al. 2019; Marneweck et al. 2018; Parikh et al. 2020). If a lack of anticipatory control of digit placement as it happens due to aging (Parikh and Cole 2012; Rao et al. 2021) or neurological conditions such as Parkinson’s disease (Lukos et al. 2010) or cerebral palsy (Gutterman et al. 2021) goes uncompensated through appropriate modulation in digit forces, then a failure occurs in achieving the task goal. Gutterman et al. observed that neurotypical children 8.7–13.9 years of age anticipated and modulated digit placement over 10 successive trials, but were unable to concurrently generate digit forces to achieve a target compensatory torque to minimize object roll (Gutterman et al. 2021). Notably, the sensorimotor memories for digit placement and digit force control may be formed independently (Lukos et al. 2008; Marneweck et al. 2018), thus raising the possibility of distinct developmental trajectories. The age at which the anticipatory control of digit position develops in children remains to be known, limiting our ability to comprehensively characterize the development of dexterity.

This study aimed to determine the development of anticipatory control of digit position for successful object manipulation in children aged 5–17 years using a paradigm that has been well-studied in adults. We divided participants into three age groups: 5–7 years of age, 8–10 years of age, and 15–17 years of age. They were instructed to learn to lift an inverted T-shaped object with an off-centered mass straight upward from a table. The object, made to resemble a rocket, tilts when lifting it unless participants use anticipatory control of digit position and forces to counteract the torque imparted on the object due to the off-centered mass. We expected that 8–10- and 15–17-year-old children would learn to modulate their digit positions, based on a study that used a similar grip device (Gutterman et al. 2021). For learned trials (those where participants achieved the target torque by covarying digit placement and forces to minimize tilt), we expected that children 15–17 years old would exhibit a stronger force-to-position strategy than 8–10-year-olds. Children 5–7 years old are still developing digit force control, and because digit force coordination is still developing, we did not expect that they would learn to modulate their digit positions or to covary their digit placement and digit forces.

Methods

Participants

Thirty-nine typically developing children provided assent and a parent provided written informed consent to participate in this study. Participants were assigned to three groups based on their age. Group 1 were 5.08–7.91 years old (n = 13, mean 6.56 ± 0.99 years); Group 2 were 8.30–10.90 years old (n = 15, mean 9.38 ± 0.98 years); and Group 3 were 15.16–17.24 years (n = 11, mean 16.22 ± 0.64 years). Subject characteristics are presented in Table 1. The study required each participant to participate in one session lasting approximately 2 h. Participants were right-hand dominant with normal or corrected-to-normal vision, no upper limb injury, and no musculoskeletal or neuromuscular disorders as reported by a parent. The study procedures were approved by the Institutional Review Board of the University of Houston.

Table 1.

Group characteristics

	Group
	1	2	3
n	13	15	11
Age	6.56 (0.99)	9.38 (0.98)	16.22 (0.64)
Gender	6M/7F	8M/7F	5M/6F
Hand length (mm)	138.5 (8.1)	158.9 (7.6)	188.9 (12.9)
Hand span (mm)	156.7 (9.5)	174.1 (13.2)	206.4 (21.3)
Pinch strength (Ib)	2.5 (0.6)	3.2 (0.2)	5.4 (1.3)
Key pinch strength (Ib)	3.9 (0.7)	4.6 (1.1)	7.5 (1.1)
Stereognosis score 1–9,	7 [3], 8 [3],	7 [0], 8 [5],	7 [0], 8 [0],
[n]a	9 [7]	9 [10]	9 [11]

Mean (SD), ^aScore [count]

The age of 5 years was chosen as the youngest age, as this is typically the preschool and kindergarten years when fine motor skill development such as writing and using precision grip becomes more important. Developmental delay in fine motor skills is typically identified near this age. The age range of 6–10 years was chosen based on earlier work reporting this range as when adult-like anticipatory control of digit forces develops (without considering digit placement; Forssberg et al., 1991; Forssberg et al., 1992; Gordon 2001). The age range of 11–14 years was not chosen because a study with a similar grip device has examined digit coordination in this age group (Gutterman et al. 2021). Children 15–17 years were chosen based on literature reporting the development of dexterity into these years (Dayanidhi et al. 2013; Eliasson et al. 2006), but with cortical connectivity and performance similar to that in adults (Beck et al. 2021).

Procedures

Digit force and position characterization:

A custom-designed inverted T-shape device (Parikh et al. 2014, 2020; Rao and Parikh 2019) made to resemble a toy rocket was instrumented with two 6-axis force and torque transducers sampled at 1000 Hz. (Fig. 1; Nano-25, ATI Industrial Automation, Garner, NC). The device measures digit forces and torque both normal and tangential to the graspable surfaces, which are covered in sandpaper (grit #320). The base consists of three compartments (left, center, and right). A 200-g mass is inserted in the left (LCM) or right (RCM) bottom compartment to shift the mass distribution to the left or right of the vertical midline, respectively. LCM or RCM mass generates external torque $({\mathrm{T}}_{\text{ext}})$ of ± 122 N·mm following lift onset. Mass distribution remains symmetrical when added to the center (CCM). To prevent visual identification of the location of the mass, the view of the base compartments is blocked by a cover. The total weight including sensors and hidden mass is 421.9 g, which children as young as 3–4 years can lift using their index finger and thumb (Forssberg et al., 1991; Forssberg et al., 1992; Gordon et al. 1993). The vertical plates are separated a total width of 5 cm, and the height of the plates is 10 cm. An electromagnetic sensor (Polhemus FASTRAK; 0.05° resolution, 120 Hz) is attached inside the rocket cone to measure the object tilt, or roll. The roll of the object is defined as the angle between the gravitational vector and the vertical axis of the grip device, contained in the frontal plane of the device. Force sensor data is recorded through an analog-to-digital converter board at 1 kHz (PCI-6220 DAQ, National Instruments, Austin, TX). Custom software (LabView, NI) synchronizes all sensor data collected from the grip device.

Fig. 1.

A Frontal inner view of the custom grip device allowing choice of position of thumb and index finger along two vertical plates mounted onto 2 force/torque sensors (a) mounted on a central plate. A position tracker (b) is mounted inside the cone. A mass (c) is added into one of 3 compartments at the base. B Free body diagram with hidden mass in the LCM position (Left with respect to subject). C Photograph of the grip device with the front cover

Object manipulation task

All participants had a full view of their right hand and the grip device during the task. The participant’s hands rested on a table placed directly in front of them to ensure that the start position and arm posture were consistent throughout the experiment. The grip device (Fig. 1) was positioned 20 cm from the edge of the table. Participants were instructed that they could grasp anywhere along the vertical plates to perform the task and to lift the object straight upward to a height of ~10 cm from the table as demonstrated without letting it tilt. Participants were instructed about a change in CM of the object after each block. However, they had no a priori information about the object CM's location. During each trial, participants attempted to lift the object as straight as possible off the table, that is, to prevent the object from rotating in the frontal plane due to the asymmetrical mass distribution. To be successful, participants had to exert a compensatory torque $({\mathrm{T}}_{\text{com}})$ of the same magnitude but in the opposite direction of ${\mathrm{T}}_{\text{ext}}$ in an anticipatory fashion—at object lift onset.

After the demonstration, each participant performed two practice trials with the centered mass (CCM). Corrective feedback was given if the participant used additional digits or otherwise did not grip the device as instructed, or if the object tilted rather than lifting straight vertically. Before each block, participants were reminded that the rocket may tilt in either direction. A block of 10 trials each was performed for LCM, CCM, or RCM conditions. Between each block, participants turned away from the device and rested for 30–60 s while the mass location was changed out of their view. Although participants could not anticipate CM location at the beginning of each block of trials (i.e., trial 1), they were informed that no other changes were made for subsequent trials in that block of trials. They were also reminded between blocks that their goal was to lift the object straight off the table. To minimize object roll, participants were required to produce a torque on the object before lift onset equal in magnitude but opposite in direction to ${\mathrm{T}}_{\text{ext}}$ generated by the added mass (target torque = −122 N·mm for RCM; + 122 N·mm for LCM). Participants had to anticipate, rather than react to, the external torque through consecutive lifts. After the lift, the resulting object roll provided visual feedback about the extent to which the task was correctly performed.

Experimental design

The experiment was conducted in a well-illuminated and quiet room. Participants began with a hand preference test, and their hand span and length were measured (Thomas et al. 2022). For strength measures, participants were instructed to pinch a standard pinch gauge (accuracy ± 1%, B&L Eng, CA) as hard as possible using a precision grip and key grip. We used the highest score of two attempts for our analysis. For stereognosis, three familiar objects (key, spoon, pin) and six similar matched objects (e.g., button/coin; paper clip/safety pin; pen/pencil) were placed into the right hand from behind a screen, not visible to the participant. Participants then verbally identified or matched each object to a visual display of the same objects, with scores ranging from 0 to 9 (Sakzewski et al. 2010). Participants then performed the object manipulation task over 3 blocks of 10 trials each (CCM, LCM, and RCM) after 2 practice trials. The order of placement of the mass at the start of each block of trials was pseudorandomized and counterbalanced. To lift the object without tilting it, the torque that is exerted onto the apparatus must be equal in magnitude but opposite in direction to the external torque generated by a hidden off-centered mass. This means that participants had to anticipate the external torque through consecutive lifts.

Data analysis

Force/torque measures

The data were run through a 5th order low pass Butterworth filter (Fu et al. 2010). Position data were resampled at the same rate as the force data (MATLAB; MathWorks). Force, torque, and object roll data were used to compute the following variables (illustrated also in Fig. 1B): (1) Digit load force (LF): the vertical force component parallel to the grip surface produced by each digit to lift the object. LF was then used to compute the ${\mathrm{d}}_{\text{LF}}$, defined as the difference between the thumb and index finger load forces. The digit normal force $({\mathrm{F}}_{\mathrm{N}})$ is the force component perpendicular to the grip surface, and the grip force $({\mathrm{F}}_{\text{GF}})$, was defined as the average of the normal forces produced by each digit. (2) Digit center of pressure (CoP) is defined as the vertical coordinate of the point of resultant force application by each digit on the grip surface, calculated using the following equation (relative to the sensor’s frame of reference; z is the graspable plate thickness):

$$Co{P}_y=[Mx-(Fy\times z)]÷F_N$$ (1)

Using these variables, compensatory torque $({\mathrm{T}}_{\text{com}})$ generated by the subject is computed as:

$$T_{com}=\left(\frac{w}{2}\cdot d_{LF}\right)+(d_y\cdot F_{GF})$$ (2)

where ${\mathrm{d}}_{\text{LF}}$ is the difference between the LF of the thumb and index finger, ${\mathrm{d}}_{\mathrm{y}}$ is the vertical difference between the CoP of the thumb and index finger, and $w$ is the grip width (Davare et al. 2019; Parikh et al. 2020). The task goal of minimizing object tilt requires subjects to apply a torque on the object of similar magnitude but opposite in direction. ${\mathrm{T}}_{\text{com}}$ is computed at the time of object lift onset to quantify anticipatory control of manipulation (Davare et al. 2019; Forssberg et al., 1991; Forssberg et al., 1992; Fu et al. 2010; Parikh et al. 2020; Parikh and Santello 2017). This is the time before subjects perceive and react to the ${\mathrm{T}}_{\text{ext}}$. Object lift onset is defined as the time at which the vertical position of the grip device crosses and remains above a threshold (mean+2SD of the baseline) for 200 ms (Zhang et al. 2010). (3) Peak object roll is computed as the maximum roll occurring ~150–200 ms after object lift-off. This was confirmed by inspecting object roll for each trial. Erroneous anticipatory control of digit position and forces results in object roll at this time, before corrective responses to counter roll can be made at reaction time latencies (Lukos et al. 2007).

Statistics

To ensure that performance variables using the custom-designed inverted T-shape device (i.e., the “rocket”) are comparable to those used in studies of dexterous object manipulation in adults, we first looked at the key performance variables of ${\mathrm{T}}_{\text{com}}$ and Peak roll in the adolescent group. Adolescents aged 15–17 years were considered as a control group to verify that the new device and protocol would show the same trends in ${\mathrm{T}}_{\text{com}}$ and Peak roll across 10 blocked trials as in numerous studies with young adults (Davare et al. 2019; Fu et al. 2010; Parikh et al. 2020; Zhang et al. 2010). To assess learning-related changes, we performed a within-subject repeated-measures ANOVA with “CM” (3 levels: RCM, LCM, CCM) and “Trial” (10 levels) as factors. Post hoc comparisons were performed on neighboring trials using paired t-tests with Bonferroni corrections. From this analysis, we determined that the difference between the 1st and the mean of the last 5 trials (consistent with literature; Parikh et al. 2020) was appropriate for our group-wide comparisons of performance “pre” and “post” learning. Because ${\mathrm{T}}_{\text{com}}$ and Peak roll results were identical (also consistent with literature; Parikh et al 2020), we focused on ${\mathrm{T}}_{\text{com}}$ and the behavioral components (e.g., digit separation dy) for group-wide analyses.

We then compared the performance of all groups using a between-within repeated measures ANOVA with Group (1–3) as the between-subject factor and “CM” (3 levels) and “Trial” (2 levels–1st trial and mean of last 5 trials) as within-subject factors. We also considered how differences in hand size or grip strength may have played a role in the performance of the task and thus we included these measures as covariates as statistically appropriate. We found that all measures were highly correlated with each other (Table 2). Thus, to select which covariate to use, we inspected plots for normality, linearity, homogeneity of variance, and tested for homogeneity of regression on each variable to ensure assumptions were met for inclusion. When differences were found in learning “Trial” or between groups, post-hoc t-tests were computed and adjustments for multiple tests maintained a nominal type I error rate (α = 0.05). This same procedure was repeated for the components of ${\mathrm{T}}_{\text{com}}$ (${\mathrm{d}}_{\mathrm{y}}$, ${\mathrm{d}}_{\text{LF}}$, and ${\mathrm{F}}_{\text{GF}}$) to investigate differences in how ${\mathrm{T}}_{\text{com}}$ was generated. These analyses were important to determine if digit separation or digit force generation was different between groups and changing from the 1st to last 5 trials. We also used independent t-tests to compare the SD of ${\mathrm{d}}_{\mathrm{y}}$ between groups, to investigate the variability in digit placement across groups. Statistical analyses were performed using SPSS software version 28.0 (IBM, USA).

Table 2.

Correlations between hand size and grip strength

Pearson correlations
	Span	Length	Pinch
Length	0.908**
Pinch	0.812**	0.802**
Key	0.741**	0.800**	0.834**

^**Correlation is significant at the 0.01 level (2-tailed)

p-values based on 1000 bootstrap samples results

Results

All participants completed all measures of the study without requiring any breaks or reporting any adverse effects.

Groups 1 and 2 did not apply the correct torque over 10 repeated trials with off-centered mass

To compare ${\mathrm{T}}_{\text{com}}$, we found that hand length was the most suitable measure to use as a covariate. After controlling for hand length, we found a difference between initial and later-trial ${\mathrm{T}}_{\text{com}}$ (main effect of Trial: F_1,35 = 5.046, p = 0.031, n_p² = 0.013), and that ${\mathrm{T}}_{\text{com}}$ at each CM condition differed according to Group (Group x CM interaction effect: F_4,70 = 2.711, p = 0.037, n_p² = 0.057; see Fig. 2).

Fig. 2.

${\mathrm{T}}_{\text{com}}$ at the RCM and LCM conditions. Groups 1 and 2 did not reach the target torque (dashed line) in either RCM or LCM condition

We then computed between-group post hoc independent t-tests and found that for all CM conditions, there were no differences in the initial ${\mathrm{T}}_{\text{com}}$. However, for later trial ${\mathrm{T}}_{\text{com}}$, we found significant differences between Group 1 and Group 3 in both LCM and RCM conditions (RCM: t₂₂ = 5.720, p < 0.001; LCM: t₂₂ = −8.804, p < 0.001; corrected α = 0.008). We found a significant difference between Group 2 and Group 3 in the LCM condition (t₂₄ = −6.773, p < 0.001; corrected α = 0.008), but not in the RCM condition (t₂₄ = 2.756, p = 0.011; corrected α = 0.008). No group differences were found in CCM later-trial ${\mathrm{T}}_{\text{com}}$ (all p values > 0.05).

Anticipatory control of digit placement

The adolescents modulated their digit placement depending on the CM condition, similar to young adults. In the LCM condition, their digits were collinear (Trial 1), and after a few trials, their thumb was placed ~10 mm higher than the index finger (Fig. 3). As this separation occurred, a slightly greater load force was produced by the thumb. In contrast, Group 1 began their trials with a higher thumb with minimal change in digit position across the 10 trials. This group failed to concurrently distribute the load forces between their index finger and thumb to apply more load force at the thumb in relation to the index finger (Fig. 3). Similarly, Group 2 showed a small change in digit placement across the trials with minimal or no compensation in the load force distribution between the thumb and index finger.

Fig. 3.

Group 1 (red), Group 2 (blue), and Group 3 (black). Top: Mean (SE) dy is plotted across the 10 trials in LCM and RCM conditions. Positive values indicate the thumb was placed higher than the index, Negative values indicate the index was placed higher than the thumb, Values at zero indicate the digits were placed collinear. Differences were found between Group 1 and Group 3 for LCM and RCM. Bottom: Mean (SE) ${\mathrm{d}}_{\text{LF}}$ is plotted across the 10 trials in LCM and RCM conditions. Positive ${\mathrm{d}}_{\text{LF}}$ indicate more thumb load force, negative indicates more index finger load force. Values near zero indicate equally distributed load force

In RCM, Group 3 placed the thumb lower than the index finger and this separation increased throughout trials 6–10. Importantly, they were able to compensate to reach the target ${\mathrm{T}}_{\text{com}}$ and achieve stable performance in trials 6–10 (Fig. 2) by distributing more load force to the index finger (negative ${\mathrm{d}}_{\text{LF}}$, Fig. 3).

It is possible that the younger groups did not separate their digits to create a mechanical advantage because their hands were smaller, or their grip strength was lower. Due to the mechanics of the hand and fingers, it might be more difficult for a smaller hand to transmit force with the index finger when placed higher than the thumb (RCM). Controlling for hand length, we found that digit separation at each CM was different between groups (Group x CM interaction: F_2,70 = 4.113, p = 0.020, n_p² = 0.105) and that this difference by CM condition may also have depended on Trial (Group x CM x Trial: F_4,70 = 2.507, p = 0.050, n_p² = 0.125).

Post hoc analysis (Fig. 4) found that for the LCM condition, there was not a significant change in dy from initial to later trials in the youngest group (Group 1: 2.73 mm, t₁₂ = −2.243, p = 0.045; corrected α = 0.005). However, Group 2’s difference of 5.55 mm and Group 3’s initial-later difference of 12.63 mm were significant (Group 2: t₁₄ = −3.700, p = 0.002; Group 3: t₁₀ = −5.024, p < 0.001; corrected α = 0.005). At RCM, there was no change in ${\mathrm{d}}_{\mathrm{y}}$ for the younger groups (Group 1: 3.33 mm, t₁₂ = 2.494, p = 0.028; Group 2: 4.09 mm, t₁₄ = 2.190, p = 0.046; corrected α = 0.005), while Group 3’s ${\mathrm{d}}_{\mathrm{y}}$ change was significant (7.80 mm, t₁₀ = 5.105, p < 0.001; corrected α = 0.005). At CCM, there were no differences between initial and later ${\mathrm{d}}_{\mathrm{y}}$ in any of the Groups (all p values > 0.005).

Fig. 4.

Change in digit positioning (dy) Group 1 (red), Group 2 (blue), and Group 3 (black). For each CM, grouped bars indicate the mean (SE) initial trial and later trial dy for each group

For ${\mathrm{d}}_{\text{LF}}$ we found no main or interaction effects (all p values > 0.05). In general, when the mass was on the left, more load force was applied by the index finger. When the mass was on the right, more load force was applied by the thumb. This was observed most clearly in Group 3 (Fig. 3).

In the ${\mathrm{F}}_{\text{GF}}$ analysis, we found no main or interaction effects after controlling for differences in pinch strength (Fig. 5). These results suggest that a partial explanation for why the younger groups did not reach target ${\mathrm{T}}_{\text{com}}$ was because their lower strength prevented them from applying enough ${\mathrm{F}}_{\text{GF}}$. However, Group 3 (successful at reaching target ${\mathrm{T}}_{\text{com}}$) produced more ${\mathrm{F}}_{\text{GF}}$ at RCM than at LCM. This suggests that in addition to producing more ${\mathrm{F}}_{\text{GF}}$, younger groups would also have needed to modulate ${\mathrm{F}}_{\text{GF}}$ based on CM condition.

Fig. 5.

Later trial ${\mathrm{F}}_{\text{GF}}$ mean (SEM) for Group 1 (red), Group 2 (blue), and Group 3 (black) for all three CM conditions. We found that Group 3 used more ${\mathrm{F}}_{\text{GF}}$ at RCM than at CCM, while the other groups ${\mathrm{F}}_{\text{GF}}$ did not vary by CM

Discussion

Our study examined the development of the ability to form and retrieve sensorimotor memories (i.e., anticipatory control) of digit positions to apply accurate torque on an object for minimization of roll in typically developing children. We found that children aged 15–17 years of age demonstrated anticipatory control of digit positions along with significant digit force-to-position coordination resulting in the minimization of object roll. Children aged 5–7 years when compared with the adolescents group failed to use sensorimotor memories from previous trials about the object’s CM location to plan their digit positions and digit forces resulting in an inaccurate torque application causing significant object roll. Children aged 8–10 years when compared with the adolescents group showed a modest ability to use sensorimotor memories from previous trials.

Children 5–7 years did not demonstrate anticipatory modulation of digit positions

Learning to generate compensatory torque before the time of lift-off requires that a relation is learned between digit positions and forces. Raising the height of the index finger relative to the thumb for the RCM condition and vice versa for the LCM condition increases the moment arm so that the digit forces to lift the object without letting it tilt are more evenly distributed between the digits. Adults consistently choose this strategy to successfully minimize object roll (Fu et al. 2010; Lukos et al. 2008; Parikh et al. 2020). This strategy may be more efficient in terms of energy cost, avoid potential injury caused by excessive force at a single digit (Fu et al. 2010), and may partially relieve the need to adopt different force-sharing patterns between individual digits (Lukos et al. 2007). The 15–17-year-olds learned to lift the object without a tilt by modulating their digit positions and forces. That is, with repeated lifts, these children were able to scale their digit position to counteract the external torque exerted on the object created by the off-centered mass. Younger children in our study either did not choose this digit positioning strategy (5–7-year-olds) or chose this strategy in one CM condition and not the other (8–10-year-olds). Several possible factors may have contributed to this result, preventing them from forming or retrieving the sensorimotor memory for grasp kinematics. However, none of the covariates, such as hand length and pinch strength, contributed to this effect.

Digit interdependence

For the modulation of digit position to take place, the thumb and index finger must abduct/adduct and flex/extend relative to the other digit. Such an ability might depend on the development of independent control of movement at each digit (Schieber and Santello 2004). Digit independence increases throughout childhood (Gordon 2001; Shim et al. 2007). The thumb and index finger are the two most individuated digits in adults (Lukos et al. 2007; Schieber and Santello 2004). When tasked to generate maximum flexion force of the index finger, children 5–9 years old could only do so by also flexing the middle, ring and little fingers, suggesting that the ability of the index finger to exert force independently may not develop until the age of 10 years (Shim et al. 2007). A similar developmental pattern for digit independence has been observed at submaximal forces (McCall et al. 2023). In our study, the 5–7-year-olds may not have fully developed the ability to independently control their digits, thus showing a lack of modulation of digit position despite repeated lifts. On the other hand, the independent control of digits may still be developing in the 8–10-year-olds because these children were able to modulate their index finger and thumb for the LCM condition but not for the RCM condition.

Anticipation of grip orientation and limb posture

The lack of modulation of digit positions may have been related to the development of mechanisms for anticipatory planning of whole hand grip posture and orientation as studied within the framework of “end state comfort” control (Wunsch and Weigelt 2016). Adults choose a grip orientation at object contact (e.g., grasp with thumb up or thumb down) to maximize “comfort” at the anticipated ending position and orientation of the object (Rosenbaum et al. 2012). “Comfort” is taken to mean that the posture of the hand and proximal joints end in a more advantageous position after the task (Rosenbaum et al. 2012). Studies of typically developing children report somewhat conflicting results regarding timeline, but overall agree that grip orientation planning develops gradually from the age of 3–10 years (Jongbloed-Pereboom et al. 2013; Wunsch and Weigelt 2016).

Children younger than 5 years selected a simplified, “default” grip orientation even with repeated attempts. They selected a grip according to the immediate task demand (lift the object) rather than according to how the object will be manipulated in future steps (Wunsch and Weigelt 2016). Children 5–10 years old are progressively better at anticipating grip posture outcomes and matching them to a desired end effect (Jongbloed-Pereboom et al. 2013; Wunsch and Weigelt 2016). We found that children 5–7 years consistently chose a default digit placement strategy over the course of 10 trials in the LCM and RCM conditions, and that children 8-10yrs were moderately successful at modulating digit positions in one of the CM conditions.

It is debatable whether direct connections can be made between the development of the “end state comfort effect” for anticipatory planning of grip orientation and digit position during our task, restricted to only one grip with little change in grip orientation required. However, the behavior in our youngest group seemed to mirror the behavior seen in the first stage of development of the “end-state comfort effect” (Wunsch and Weigelt 2016). A default “grip”, or digit positioning strategy, persisted despite experience and no correction was made to the desired action effect of minimizing object roll. The development of perceptual and/or cognitive mechanisms (i.e., gathering the relevant information and/or processing the information adequately; Ossmy et al. 2022) responsible for development of the end-state comfort effect observed in children 5–10 years old may overlap with those responsible for anticipatory digit modulation (absent in the youngest group and partial success in the middle group (8–10yrs).

Sensorimotor feedback

Consistent with precision force control literature (Forssberg et al., 1992; Forssberg et al., 1991; Gordon 2001), both younger groups (5–7 and 8–10 yrs) demonstrated anticipatory control of digit forces, as evidenced by their ability to produce grip and load forces by lift-off. However, they did not demonstrate the ability to use sensory information obtained from previous trials to modulate their digit positions and load forces in the LCM (youngest group) or in the RCM condition (both younger groups). Children 8–10 years behaved similarly to children 8.7–13.9 yrs in Gutterman et al. who observed a digit separation based on CM condition with a higher target torque (± 160 Nmm). However, they similarly did not create the separation and digit force modulation to the degree needed to generate the target compensatory torque at lift off (Gutterman et al. 2021).

A symmetrically shaped object is expected to have a symmetrical weight distribution. In our study, the CM location was predictable from trial to trial, and no explicit visual cues about the CM location were provided. We found that children 15–17-years-old initially choose to place their thumb and index finger in a collinear fashion. With repeated lifts and through sensory information, they learned to modulate their digit placement. Similar learning to anticipatorily control the digit position as per implicit cues about the object CM location has been widely reported in neurotypical adults (Fu et al. 2010; Lukos et al. 2007; Parikh et al. 2020). In contrast, our youngest group (5–7 years old) consistently used a default digit placement (no significant difference between first and last 5 trials ${\mathrm{d}}_{\mathrm{y}}$), indicating their inability to predict the CM location over blocked trials. It is possible that the symmetrical visual appearance of the object at the beginning of a trial either prevented the formation of sensorimotor memory for the modulation of digit position or interfered with the retrieval of formed sensorimotor memory.

Additionally, studies in adults have also shown that the increased uncertainty and variability of digit contact points (by allowing a choice of contact points on the object such as in our object manipulation task) shifts the control mechanism from a predominantly sensorimotor memory-based feedforward control to one that incorporates online or “real time” sensorimotor feedback at object contact (Davare et al. 2019; Parikh et al. 2020). Our younger groups had several participants who performed poorly (scores of 7/9 or 8/9) on the stereognosis test, which measures the integration of tactile and proprioceptive signals from hand and digit contact with an object, conveying information about the object’s shape at those points of contact (Sobinov and Bensmaia 2021) (Sakzewski et al. 2010). Likely, children 5–10 years old may not have developed the ability to incorporate or integrate online or “real time” sensorimotor feedback at object contact for anticipatory control of digit position for dexterous manipulation.

Anticipatory control of digit position develops later than digit forces in children

Anticipatory control of digit forces has been shown to emerge in children as young as 2 years when the grip forces and load forces during the loading phase—the phase from object contact to object lift- were shown to increase more linearly than at previous ages (Forssberg et al., 1991). From 2 to 8 years, grip and load force rate profiles progressively appear more bell-shaped as in adults, indicative of a shift from feedback-based control to feedforward anticipatory control (Forssberg et al., 1991). At the ages of 8–11 years, children can scale their load force rates in response to changes in weight of the same grip device similarly to adults (implicit learning; Forssberg et al., 1992). Similarly, in dexterous manipulation studies—absent any verbal or explicit visual cues- participants learn to anticipate their digit forces by way of both visual and somatosensory feedback at lift-off, i.e., “implicit” knowledge gained over the course of repeated trials.

Digit force planning relies on implicit learning, whereas anticipatory digit positioning can also be learned from explicit cues (Lukos et al. 2008; Zhang et al. 2010). Thus, anticipatory mechanisms for grasp kinetics (force control) and kinematics (position control) can occur separately (Lukos et al. 2008). Adolescents, like adults, used implicit learning to anticipate digit force in parallel with digit positions. However, children 5–7 years old showed no implicit learning for anticipatory digit position modulation. For children 8–10 yrs, this learning was only observed in the LCM condition and not to the extent that allowed success in minimizing object roll. Our results suggest a separate, more prolonged timeline for the development of anticipatory mechanisms of digit kinematics than for digit kinetics in children. We also extend and support the evidence that dexterity continues to develop into adolescence (Dayanidhi et al. 2013; Fuelscher et al. 2021; Wolff 2023).

Limitations

In this study we used an inverted T-shape apparatus in a laboratory setting to investigate digit positioning and digit forces at the time of object lift. The apparatus was scaled down from the sizes of objects used for this purpose in adult studies. The weight of 400 g was chosen such that the object roll could be easily detected at lift off and necessitate a finger separation strategy, while still being light enough for younger children to lift with their fingers. In real life settings, children encounter objects of many different sizes and weights, and may select different grip strategies depending on the size and weight of the object. This study is not immediately generalizable to all such conditions; however, the intent is to contribute to understanding the emergence of digit position control to inform studies of pediatric hand function. It may also help to define the most effective timing, dosage, or related considerations for interventions.

Conclusion

Our findings highlight differences in the development of anticipatory control of digit positions over the ages of 5–10 years and 15–17 years when digit positions are allowed to vary on a dexterous manipulation task. The current study supports the framework of continued development of dexterity in adolescence.

Data availability

Data that support the findings of this study have been deposited in Mendeley Data: Rose, Vivian (2024), “Anticipatory Control of Digit Kinematics”, Mendeley Data, V1, https://doi.org/10.17632/fgng74pzy2.1