Training 90° Bimanual Coordination at High Frequency Yields Dependence on Kinesthetic Information and Poor Performance of Dyadic Unimanual Coordination

Abstract

Two groups of participants were trained to be proficient at performing bimanual 90° coordination either at a high (2.5Hz) or low (0.5Hz) frequency with both kinesthetic and visual information available. At high frequency, participants trained for twice as long to achieve performance comparable to participants training at low frequency. Participants were then paired within (low-low or high-high) or between (low-high) frequency groups to perform a visually coupled dyadic unimanual 90° coordination task, during which they were free to settle at any jointly determined frequency to synchronize their rhythmic movements. The results showed that the coordination skill was frequency-specific. For dyads with one or both members who had learned the 90° bimanual coordination at low frequency, the performance settled at a low frequency (≈0.5Hz) with more successfully synchronized trials. For dyads with both members who had learned the 90° bimanual coordination at high frequency, they struggled with the task and performed poorly. The dyadic coordination settled at a higher frequency (≈1.5Hz) on average, but with twice the variability in settling frequency and significantly fewer synchronized trials. The difference between the dyadic coordination and bimanual tasks was that only visual information was available to couple the movements in the former while both kinesthetic and visual information were available in the latter. Therefore, the high frequency group must have relied on kinesthetic information to perform both coordination tasks while the low frequency group was well able to use visual information for both. In the mixed training pairs, the low frequency trained member of the pair was likely responsible for the better performance. These conclusions were consistent with results of previous studies.

1. Introduction

In a study of learning 90° rhythmic coordination, Snapp-Childs, Wilson and Bingham (2015) found that the learned performance transferred from a unimanual task to a bimanual task and vice versa. In both tasks, visual information (spatiotemporal relationship between the two oscillators) was available about the coordination while kinesthetic information (body-related coordination information) was also available in the bimanual task. However, subsequently, it was shown that the common availability of the visual information was essential for transfer of the learned skill to occur. In a similar study, Bingham, Snapp-Childs, and Zhu (2018) found no transfer when only kinesthetic information about the coordination was available in the bimanual task and only visual information in the unimanual task. These results indicated that, in each case, the information learned and used for coordination was modality specific, meaning as well that the information in the two modalities was different.

Given this, if both visual and kinesthetic modalities were available to someone learning a 90° coordination in a bimanual task, one might ask whether they would learn and use both modalities or just one and if just one, which one? Zhu, Mirich, Huang, Snapp-Childs, and Bingham (2017) trained two groups of participants to perform 90° bimanual coordination with extensive practice in which visual information about the coordinated movements was present for one group and absent for the other. Kinesthetic information was available to both groups. After training, both groups were tested with and without available visual information. Participants who learned the coordination with visual information available reverted to baseline performance level when the visual information was removed. Because kinesthetic information remained available during all training and testing, the failure to exhibit the learned coordination in the absence of visual information suggested that only visual information and not kinesthetic information had been used in learning and subsequently performing the coordination. Participants trained with only kinesthetic information struggled to perform as well as the group with visual information, that is, they had to train for much longer and still could not perform as well.

In studies of learning to perform novel rhythmic coordinations, the training is nearly always kept at a low frequency (≈ 1Hz). Frequency scaling studies investigating either unimanual or bimanual performance of a learned coordination with visual information available have found that the stability of performance has decreased with increasing frequency (Bingham, Herth, & Zhu, submitted; Huang, Layer, Smith, Bingham & Zhu, submitted). That is, the learned coordination cannot be reliably or well performed at a high frequency. The evidence indicates that it is only visual information that was being used by participants in these studies (despite kinesthetic information having been available in the bimanual tasks) and that the visual information becomes unreliable at high frequencies. Presumably the relevant information becomes difficult to resolve.

An exception to the use of low frequency movement for training was made in a recent study (Huang, et al., submitted), where a group of participants was trained to perform 90° bimanual coordination at 2.5Hz with both visual and kinesthetic information available. These participants had to train for twice as long as a group who trained at low frequency (0.5Hz) to reach a comparably good level of performance. Surprising results were obtained in this study when the trained participants were tested with frequency scaling. Intuitively, performance of rhythmic coordination at high frequency is more difficult than at low frequency, and additional practice is required for training at a high frequency to attain proficiency. However, when participants who were trained and proficient at performing 90° rhythmic coordination at high frequency were tested at progressively lower frequencies, their performance became progressively less stable until reaching pre-training levels of performance at the low training frequency (0.5Hz). These participants were unable to perform at low frequency the learned coordination that they could perform at high frequency. The results of frequency scaling were the opposite for the high frequency trained participants of what they were for the low frequency trained participants. The authors of the study inferred that the latter, low frequency group, learned to use visual information to perform the 90° coordination while the former, high frequency group, learned to use kinesthetic information to perform the coordination. The inference was that kinesthetic information is easier to discriminate or resolve than visual information at high frequencies and more difficult to discriminate at low frequencies, if the coordination was entrained by high frequencies. Thus, participants trained at high frequency naturally prefer to use kinesthetic information to learn the coordination with the result that they are unable to exhibit the learned level of performance at low frequency. The counterintuitive result is that performance at low frequency is harder if you have learned to perform the coordination well at high frequency.

We now further investigated this information-modality-specific-to-frequency hypothesis by testing the participants that were trained in the previous study (Huang et al., submitted) in a between-person 90° coordination task. Participants were paired either within (low-low or high-high) or between (low-high) training frequency groups to perform a dyadic unimanual 90° coordination task. Each dyad was allowed to settle at a jointly determined frequency to achieve synchrony of their rhythmic movements at a 90° relative phase. The relevance of this dyadic coordination task to a test of the information modality hypothesis was that only visual information would be available about the coordination. Thus, if the hypothesis is correct, then low-low dyads should reliably settle at a low frequency (≈0.5 Hz) and synchronize well in good performance of 90° coordination, while high-high dyads should be relatively at sea, meaning that the settling frequency should be less reliable, and synchronization should be poorer or less frequent. Finally, the mixed low-high dyads might perform well if the “high” member of the pair allowed the “low” member to establish a low settling frequency for dyadic coordination. Otherwise, such mixed pairs should not perform well. In this dyadic 90° coordination task mediated by visual information, the low frequency trained participants should be the champions at low frequency while the high frequency trained participants should lose the ground to be able to use their learned skill. The rationale for above hypotheses is that the skill of the low frequency trained participants depends on the visual information that is available in the dyadic coordination task, while that of the high frequency trained participants depends on kinesthetic information about the coordination to which they do not have access in this dyadic task.

2. Methods

2.1. Participants

Thirty-three participants aged between 19-29 years (19 females and 14 males, mean age = 22.12 ± 2.26 years) were recruited at the University of Wyoming through a flyer. Thirty-two participants were self-reported as right-handed while one participant was left-handed. All participants had normal or corrected-to-normal vision and were naive to the experimental questions and tasks. The study had been approved by the Institutional Review Board at University of Wyoming, Laramie. Informed consent had been obtained from all participants.

The participants were previously trained to be skilled (above a criterion level of 60% PTT¹) at producing bimanual 90° coordination at a specific frequency. 16 participants (9 Females & 7 Males) were trained at 0.5Hz and 17 participants (10 Females & 7 Males) at 2.5Hz to perform bimanual 90° coordination with proficiency, before being paired to perform the dyadic unimanual coordination task.

Due to the various availability of participants for the follow-up dyadic coordination task, maximizing the pairing possibility was infeasible. Each participant returned to the lab to participate in 5 pairing sessions on average. Each trained participant was paired with about 3 other participants who had trained at the same frequency (Low-Low or High-High). Participants were also paired to about 2 others who had trained at different frequency (Low-High). This yielded 25 pairs of High-High, 23 pairs of Low-Low, and 30 pairs of Low-High.

2.2. Apparatus and Procedure

A computer-joystick system as seen in Figure 1 was used for participants to practice 90° bimanual coordination at either low or high frequency. In the dyadic task, participants were asked to wear earplugs and manually control one joystick to move one of the two dots on the screen together as the other dot was similarly controlled by the other participant in the pair. Accordingly, the two joysticks were each controlled unimanually by each of two participants sitting side by side and separated by a wall so that both participants could see the shared computer display without seeing or hearing the limb movement of the other (see setup for the task in Figure 2).

Figure 1.

An illustration of experimental apparatus. A 15” PC laptop was set on the top level of a custom-built cart, and a wooden board was set perpendicularly on the lower level with two joysticks attached.

Figure 2.

An illustration of experimental setup for dyadic task. Participants controlled one joystick to produce the dyadic 90° coordination. Seats were separated by a wall (dashed line), thus, limb movements were not visible to one another while vision of computer display was shared.

Before performing the dyadic coordination task, participants were warmed up by performing the bimanual coordination task at their respective training frequency for three blocks of 20-second trials. The audio metronome was provided in each trial to guide the movement frequency, and the performance in each trial including the produced frequency and quality of coordination (measured by PTT) were also recorded and presented to each participant. While their produced frequency was to be maintained close to the training frequency, the mean performance of the three warm-up blocks was to be at least 60% PTT before they could start the new experiment. Then, the pair of participants were randomly assigned to one joystick/seat in the dyadic condition. Without any visual or audio demonstration, both participants were asked to move the joystick up and down using the dominant hand so that the two dots displayed on the computer screen would move up and down rhythmically to produce 90° coordination. Participants were asked not to communicate with one another but to figure out how to produce the 90° coordination pattern on screen and maintain that pattern as long as possible. A total of 20 trials was performed without visual feedback or auditory metronome and each trial was 20 seconds long.

2.3. Data Analysis

Two 60Hz position time series from each trial were filtered using a low-pass Butterworth filter of second order with a cut-off frequency of 10Hz, and then differentiated using a central difference method to yield a velocity time series. These were used to compute a time series of relative phase produced by the pair, the key measure of coordination. Proportion of Time-on-Task (PTT) was then computed using the range of 90°±20°. This captured both the accuracy and stability of coordination performance.

For each moving dot in a trial, the produced frequency can be determined by calculating power spectral density (PSD) of the time series data from the joystick. Using a Fourier transform (FFT), the time series data can be decomposed into a spectrum of frequencies over a continuous range. Then, PSD computes the power (or energy) present in the data as a function of frequency. Consequently, the frequency associated with the strongest power can be identified, and then used as the produced frequency in each trial. Since there are two dots/joysticks moving in the coordination task, the produced frequency of the coordination in a trial was calculated by averaging the produced frequencies of the two oscillators. In the bimanual coordination task, both dots/joysticks were controlled by one person, and the difference between the two produced frequencies was less than 0.01Hz, so the two oscillators were synchronized. However, in the dyadic unimanual coordination task, each oscillator was controlled by a person who was exploring frequency to produce the stable 90° between person coordination. The difference between the two produced frequencies would likely be greater than 0.01Hz in some trials, until the pair found a way to synchronize the two unimanual movements to produce the coordination. Therefore, the trials in which the difference between the two produced frequencies was less than 0.01Hz were counted as synchronized trials. Then, the average of the two produced frequencies was defined as the settling frequency in the coordination. To quantify the coordination performance, the mean PTT of the synchronized trials was multiplied by the percentage of the total trials that were synchronized (number of synchronized trial/20 x 100%) to yield the synchronized PTT.

To determine the effect of previous training frequencies of the members of a pair on the coordination performance, one-way ANOVAs were performed on the mean settling frequency, the number of synchronized trials, and the synchronized PTTs, treating the paring condition as a between-subject factor. In addition, a Pearson’s correlation was performed to determine the relationship between mean settling frequencies and the number of synchronized trials.

The statistical significance level for all analyses were kept at α = .05.

3. Results

Table 1 shows the mean settling frequencies, proportion of synchronized trials, and mean dyadic PTTs for all pairing conditions: high-high (HH), low-high (LH), and low-low (LL). The one-way ANOVAs revealed significant group difference for mean settling frequencies (F_2,75 = 42.17, p < .001, ${\text{η}}_{\text{p}}^{\text{2}}=.53$), mean proportion of synchronized trials (F_2,75 = 20.38, p < .001, ${\text{η}}_{\text{p}}^{\text{2}}=.35$), and mean synchronized PTTs (F_2,75 = 11.63, p < .001, ${\text{η}}_{\text{p}}^{\text{2}}=.24$). Post hoc Tukey HSD analysis showed that the mean settling frequency of HH pairs was significantly higher than that of LH and LL pairs (both p’s < .001), with no difference detected between the latter two (p > .05). With respect to the proportion of synchronized trials, HH pairs had a significantly smaller proportion than the other two pairs (both p’s < .001), with again no difference detected between the latter two. Finally, in respect to the synchronized PTT, HH pairs were significantly (p < .001) lower than the other two pairs (LH and LL), with no difference detected between the latter two.

Table 1.

Dyadic Coordination Performance

Pairing Condition	N	Settling Freq (Hz) F2,75 = 42.17 §	Proportion of Synchronized Trials (%) F2,75 = 20.38 §	Synchronized PTT (PTT x % of Synchronized Trial) F2,75 = 11.63 §
HH	25	1.50 ± .59 a	.53 ± .25 a	.21 ± .11 a
HL / LH	30	.69 ± .23 b	.82 ± .17 b	.31 ± 10 b
LL	23	.58 ± .25 b	.88 ± .18 b	.35 ± .11 b

^§indicates significant main effect of pairing condition (p < .001)

^{a & b}indicate significant difference among pairing conditions where “a” is significantly different from “b”s (p < .001).

A significant negative correlation (r(78) = −.67, p < .001) was found for settling frequencies and number of synchronized trials as shown in Figure 3, which suggested that the dyadic coordination was more stable at the lower settling frequencies.

Figure 3.

Pearson’s correlation between mean settling frequencies and the number of synchronized trials. The correlation was significant (r = −0.67; p <.001).

4. Discussion

Previous studies have shown that perceptual-motor learning of bimanual coordination is modality-specific and visual information supersedes kinesthetic information for learning due to its salience at low movement frequencies. Given these results, we tested the effects of training a new coordination at relatively high frequency (Huang et al., submitted). We found as might be expected that twice as much training was required to achieve a comparable level of skill at the higher frequency. But the surprising results came with the response to a frequency scaling test where we found that the good skill exhibited at high frequency by those trained at high frequency evaporated at lower frequencies. How to account for this? We hypothesized that the salience of visual information, preferred at lower frequencies, would deteriorate with increase in frequency, and that kinesthetic information would be preferred when learning a new bimanual coordination at high frequency. To test this hypothesis, we now tested frequency-specific bimanual coordination training to see if the new coordination skill would continue to be exhibited in performance that was mediated strictly by visual information about coordination. We paired participants trained to perform 90° bimanual coordination at low or high frequency in different ways to see how well they would produce the unimanual 90° coordination in a visually mediated dyadic coordination task in the absence of kinesthetic information about the coordination. If visual information about 90° relative phase was learned at low frequency, and kinesthetic information about 90° relative phase was learned at high frequency, we would expect that dyads trained to perform bimanual coordination at 0.5Hz to perform the dyadic coordination task better than dyads trained at 2.5Hz. This was what we found. So, the main hypothesis tested in this study was supported by the results. However, there are other findings produced by the current study, that warrant discussion.

First, perceptual-motor learning of bimanual coordination is not only modality-specific, but also frequency-specific. The existing literature on perceptual-motor learning of 90° bimanual coordination (Kovacs and Shea, 2011; Chiou and Chang, 2016; Huang, Dai & Zhu, 2019) has reliably used a relatively low frequency (approximately 0.75Hz) to present perceptual (auditory, visual or kinesthetic) information during training to facilitate the learning. It remained unknown whether a new coordination could be acquired as well with perceptual information presented at a high frequency. Would performing 90° bimanual coordination at high frequency result in a switch to other coordinations, as has been found with 180° coordination, so that the 90° pattern could not be learned? The findings from both the current and the study by Herth et al. (submitted) showed that 90° bimanual coordination can be learned at a high frequency as well as at a low frequency if the intentional correction of errors was allowed. We found that pre-training performance was better at 0.5Hz than at 2.5Hz, which was reasonable because performing a novel bimanual coordination at high frequency is very challenging. However, participants improved the coordination performance significantly to reach equal levels of proficiency at low and high frequency so long as an ample amount of practice was allowed at the higher training frequency.

Second, the use of information for coordination could be different between the bimanual and dyadic coordination tasks. For participants both trained at low frequency, the dyadic coordination settled reliably at about 0.5Hz with a greater proportion of synchronized trials. On the other hand, when both participants were trained at high frequency, they slowed from their training frequency of 2.5Hz to 1.5Hz on average to perform the visually mediated dyadic coordination task, but their performance was equivalent to their untrained bimanual coordination performance (mean synchronized PTT ≈ 20%). Furthermore, while the average settling frequency was 1.5Hz, the standard deviation was nearly three times that for those trained and settling at ≈ 0.5Hz. The high-high pairs were all over the place in respect to settling frequency. The evidence showed that they really were unable to perform the task despite the fact that they had trained for twice as long as participants who had trained at low frequency! They failed to exhibit any of the learned skill they had acquired in their high frequency training. It was our speculation that the skill learned in the bimanual coordination task at high frequency was dependent on kinesthetic information about the coordination, which simply became unavailable in the dyadic coordination task, so no transfer of skill from bimanual to dyadic coordination was possible. Such a hypothesis could be alternatively tested by studies using visual perturbation or a complete removal of visual information to examine the bimanual coordination performance following the frequency-specific training, in which the participants trained with a high frequency might instead demonstrate a better performance.

Finally, when individuals trained at 2.5Hz were paired with those trained at 0.5Hz (the HL/LH pairs), they produced dyadic coordination performance that was equivalent to the dyads both trained at 0.5Hz. This suggested that the low frequency trained member of such pairs dominated the dyadic coordination so that the individuals trained at 2.5Hz had to slow down to coordinate with the pair member trained at 0.5Hz who must have insisted on performing at this frequency. Furthermore, if the pair member trained at high frequency was unable to use the available information, then he or she must simply have moved at the low frequency and then allowed the other low frequency trained member of the pair, who was able to exhibit the new skill using the available visual information, to establish the coordination at a low frequency. This finding again supported the hypothesis that training at low frequency yielded preferential use of visual information for coordination while training at high frequency yielded preferential use of kinesthetic information about coordination.

Recently, Huang et al. (2020) systematically studied the effect of consistency between visual and kinesthetic information on performing the intrinsic bimanual coordination patterns at both high and low frequencies. The inconsistency of information did not affect the coordination stability at low frequency, but it did significantly change the stability at high frequency with kinesthetic information being predominantly used to stabilize the coordination. Now, we extended this finding by showing that learning a novel bimanual coordination pattern was frequency-specific, and the coordination learned at low frequency was vision-based but strengthened by the kinesthetic coupling, while the coordination learned at high frequency was mainly kinesthesis-based.

5. Conclusion

In the current study, two groups of participants who were proficient at performing bimanual 90° coordination either at a high (2.5Hz) or low (0.5Hz) frequency were paired either within or between frequency groups. They were asked to perform a visually coupled dyadic unimanual 90° coordination task. The results aligned with our previous finding that perceptual-motor learning of bimanual coordination was not only modality-specific but frequency-specific, where visual information was preferred to be utilized at low frequency while kinesthetic information was utilized at high frequency. In the dyadic unimanual 90° coordination task where only visual information about relative phase was available to couple the movement, those participants who were trained at high frequency were unable to use the available information, thus the dyadic coordination was mainly controlled by those low trained pair members.

Highlights.

• Perceptual-motor learning of bimanual coordination is not only modality-specific but also frequency-specific, where visual information was preferred at low movement frequency while kinesthetic information was mostly used at high movement frequency.

• The use of information for coordination was different between the bimanual and dyadic coordination tasks, where both kinesthetic and visual information were available in the former while only visual information was available in the latter.

• In the mixed training dyad, the low frequency trained member of the dyad may dominate the dyadic coordination.