Implicit audiomotor adaptation

Sensorimotor adaptation alters mappings between motor commands and their predicted outcomes. Such remapping has been extensively studied in the visual domain, but the degree to which it occurs in modalities other than vision remains less well understood. Here, we manipulated the modality of reach target presentation to compare sensorimotor adaptation in response to perturbations of visual and auditory feedback location. We compared the extent of adaptation to perturbed sensory feedback for visual and auditory sensory modalities, and the magnitude of reach-direction aftereffects when the perturbation was removed. To isolate the contribution of implicit sensorimotor recalibration to adaptation in reach direction, we held sensory prediction errors and task-performance errors constant via a task-irrelevant clamp of sensory feedback. Seventy-two participants performed one of three experiments in which target location information and endpoint reach direction feedback were presented by loudspeakers (n = 24), headphones (n = 24), or a visual display (n = 24). Presentation durations for target stimuli (500 ms) and (non-veridical) endpoint feedback of reach direction (100 ms) were matched for visual and auditory modalities. For all three groups, when endpoint feedback was perturbed, adaptation was evident: reach-directions increased significantly in the direction opposite the clamped feedback, and a significant aftereffect persisted after participants were instructed that the perturbation had been removed. This study provides new evidence that implicit sensorimotor adaptation occurs in response to perturbed auditory feedback of reach direction, suggesting that an implicit neural process to recalibrate sensory to motor maps in response to sensory prediction errors may be ubiquitous across sensory modalities.


Introduction
To complete a successful reach towards an object, a person must first detect the object, estimate where it is relative to their head and body position, and then execute a motor command based on one's internal mapping of egocentric space (Andersen et al., 1993).Hand-eye coordination is a quintessential mechanism for tool use, everyday navigation, and even survival.However, it is not our only way to encode the extrinsic space of our surroundings.The complementary mechanism of auditory localization also allows us to detect external stimuli, both within and beyond the visual field, and it enables us to perform everyday tasks like reaching for an object in the dark, or swatting a mosquitoeven when half asleep.The extent to which such audiomotor processes share characteristics with visuomotor control processes, however, remains unresolved.
Sensory prediction errors occur when the expected sensory consequences of an intended movement do not match the observed consequences (Tseng et al., 2007), for example, when a thrown ball is blown off-target by the wind.We appear to adapt our movements to such visual sensory prediction errors in an implicit and obligatory manner (Krakauer et al., 2005;Smith et al., 2006;Taylor et al., 2014;Morehead et al., 2017;Kim et al., 2018;Leow et al., 2018;Tsay et al., 2021a), which is consistent with the notion that the neural control system for movement seeks to maintain intended and actual feedback about movements in register.If this idea is correct, we should expect that implicit sensorimotor adaptation occurs for all modalities of sensory feedback related to limb movement.Importantly, although the features of implicit motor command updates to perturbations of visual feedback have been especially well documented (Mazzoni and Krakauer, 2006;Tseng et al., 2007;Taylor et al., 2014), and reaching movements can be adapted to force-field environments that perturb proprioceptive feedback in the absence of vision (Scheidt et al., 2005;Franklin et al., 2007;Lefumat et al., 2015), it is not known whether auditory sensory prediction errors lead to implicit adaptations to motor commands for reaching.
Perhaps the most common paradigm that has been used to study sensorimotor adaptation is a visuomotor rotation task, in which a digital cursor representing the hand position is rotated with respect to the movement origin to provide perturbed feedback of reach direction.This results in sensorimotor adaptation, where reaches become adjusted in the direction opposite the cursor rotation.However, multiple processes appear to drive adaptation in standard visuomotor adaptation tasks, including explicit re-aiming and reward-related processes (Izawa and Shadmehr, 2011;Taylor et al., 2014;Leow et al., 2018).This makes drawing conclusions about implicit and implicit visuomotor recalibration difficult in this task, and limits our ability to observe an isolated contribution of sensory prediction errors.
A visuomotor "error-clamp" is a variation of a visuomotor rotation task that was developed to isolate the implicit contribution of motor learning (Morehead et al., 2017).Rather than perturbing feedback of the actual reach-direction by a fixed amount, an identical feedback direction is presented each trial in an error-clamp, regardless of the actual reach direction.Participants are instructed that the cursor direction is not under their control and should be ignored.Consequently, no explicit aiming strategies are required by participants during an error-clamp task, and potential reward effects are controlled by making it impossible for the cursor to contact the target during the perturbation.Additionally, assuming that participants follow instructions and aim directly at the center of the target, the magnitude of visually encoded sensory prediction errors received by participants remain consistent during an error-clamp task, regardless of reach-accuracy and motor performance.Removing the incentive for explicit re-aiming has been shown to attenuate motor adaptation, however.While visuomotor adaptation magnitudes of up to ~ 20-25 • are typical in response to a 30 • visuomotor rotation (Krakauer et al., 2005;Brudner et al., 2016;Rand and Rentsch, 2016;Vachon et al., 2020), adaptation magnitudes of ~ 10 • were reported for a 30 • visuomotor error-clamp perturbation (Morehead et al., 2017).
Here, we compared movement corrections made to the perturbed auditory and visual feedback of endpoint reach locations, via an errorclamp paradigm.As such, if there is an implicit neural process to recalibrate sensory to motor maps that is ubiquitous across sensory modalities, we would expect to observe adaptation in reach directions when perturbed feedback is applied, along with an observable aftereffect that persists upon knowledge of the perturbation's removal.As implicit motor learning appears to be cerebellar-dependent (Tseng et al., 2007), evidence of implicit audiomotor adaptation would suggest a capability for the cerebellum to facilitate adapted motor commands for reaching irrespective of the modality of sensory prediction errors.

Participants
Datasets from seventy-two right-handed participants were collected and analyzed from the University of Queensland Department of Psychology testing pools (59 female, ages 17-32, mean age 21.6 ± 3).No participants had participated in a visuomotor reaching task previously, and participants reported no neurological disorders, as well as normal or corrected to normal vision and hearing.Participants received course credit or monetary reimbursement upon study completion.The study was approved by the Human Research Ethics Committee at The University of Queensland.All participants provided written informed consent.

Study design
Three experimental groups were tested.Twenty-four participants were allocated to the "loudspeaker group," twenty-four participants were allocated to the "headphones group," and twenty-four participants were allocated to the "visual group."All groups completed the following blocks: familiarization, baseline, perturbation, and no-feedback, described later.

Task
The task for all participants was to reach towards a transiently presented visual or auditory stimulus.Using custom Matlab software, reaching behavior was recorded with a Wacom digitizing tablet and digitizing pen at 100 Hz (Intuos PTH-851/K1-CX).The experiment began with a familiarization block consisting of 10 reaches to each of seven target locations (− 60  , 60 • ) in random order.When a participant placed the digitizing pen in the starting position for 500 ms, a "target stimulus" was presented from one of the target locations.Participants were instructed to wait for the 500 ms target stimulus to complete its presentation prior to initiating movement, and then quickly reach with a center-out ballistic straight-line motion for approximately 12 cm directly towards the perceived location of the target stimulus.For the auditory conditions, this ensured there was no overlap between the presentation of the target sound and the feedback sound.Thus, for all experimental conditions, participants were tasked with moving to a remembered location.Reach-angles were calculated once the digitizing pen traveled 8 cm from the starting position across all experimental blocks and conditions.Participants were reminded to wait for the expiration of the target when necessary, and no form of feedback was provided during the familiarization block.If a reach took longer than 300 ms to travel 8 cm, participants heard a voice message that said "too slow."Reach durations exceeding 300 ms triggered "too slow" warnings across all experimental blocks and groups.Following the initial familiarization block, there were 106 "too slow" warning trials for the visual group (0.01 % of total trials), 69 "too slow" warning trials for the loudspeaker group (0.006 % of total trials), and 61 "too slow" warning trials for the headphones group (0.005 % of total trials).None of these trials were removed from analysis.The starting hand-position was located at the bottom of a U-shaped plastic cradle fixed to the midline of the digitizing tablet, 5.5 cm from the tablet's bottom boundary (Fig. 1).Participants were instructed to hold the digitizing pen with a natural but consistent precision grip throughout the experiment.Before the experiment began, participants completed 3 practice reaches to random locations to ensure they understood how to move at the proper speed, while successfully navigating to the starting position cradle by touch.No form of online or endpoint feedback was provided during the familiarization block.
Upon completion of the familiarization block, participants started the baseline block, where they reached toward one target location presented directly in front of the body's midline along the midsagittal plane for the remainder of the experiment.This single-target "baseline" block consisted of 80 reaches.During the baseline block, a "non-veridical endpoint feedback stimulus" was presented at the target location if the hand traveled a distance of 8 cm in less than 300 ms.From the first baseline trial, participants were informed that the feedback did not reflect motor performance.Following the baseline block, participants were exposed to a perturbation block in which the "target stimulus" was presented from the same target location as the baseline block.Instead of the "non-veridical endpoint feedback stimulus" coming from the same target location however, it was presented at a clamped-angle rotated by 40 • , regardless of motor performance (adjusted either clockwise or counterclockwise of the target location for each participant, and allocated pseudo randomly).For each experimental group, 12 participants experienced a "perturbed location" in the clockwise direction, and 12 subjects experienced a perturbed location in the counterclockwise direction (also allocated pseudo randomly).In total, there were 240 trials with task-irrelevant clamped sensory feedback presented during the perturbation block.Preceding the perturbation block was a one trial B. Miller-Mills et al. demonstration of the 40 • error-clamp, and an optional eyes-closed stretch break was offered after the first 80 perturbation trials.Finally, a block consisting of 160 "no-feedback" trials was performed, in which the "target stimulus" was presented from the same target location as the baseline block, but the "non-veridical endpoint feedback stimulus" was not presented.
All participants used the same digitizing tablet and pen with the same starting position cradle.Participants were tested in the same room with the same adjustable chinrest for head posture stability while sitting.Head position was forward facing, positioned along the midsagittal plane directly in front of the baseline target locationand held consistent across experimental conditions.Participants experienced equal head movement restrictions and break opportunities.The starting hand-position was held at a similar distance from the body for participants across groups, with individual difference due to body size.Participants wore disposable rubber gloves and were instructed to keep their left hand in their lap during the entire experiment.Video recordings made at 60 FPS (with 4k resolution) were used to document auditory stimuli presentation latencies for both the loudspeaker and headphones conditions.
Task instructions.For all experimental groups, the first trial of the perturbation was experienced by the participant and then explained by the experimenter.Instructions were given that the endpoint feedback stimulus was now coming from a new and different location, and that this would continue to be the case for the remainder of the experimental block − regardless of the participant's performance.Participants were then instructed to "ignore the irrelevant endpoint feedback and continue reaching directly towards the target."Similarly, prior to the first reach of the post-perturbation no-feedback block, participants were instructed to continue reaching directly towards the target − even though they would no longer be experiencing the endpoint feedback.

Loudspeaker group
A custom loudspeaker array was designed for the experimental loudspeaker group.Seven loudspeakers (PMT-40N25AL01-04, 4.39 Ohms Full Range Speakers 5 W 100 Hz ~ 20 kHz) were spaced apart by 20 • at a distance of 150.1 ± 1.1 cm from the participant (variation due to individual differences in participant head size).Loudspeakers were 3 cm in diameter.The center of each loudspeaker was set to a height of 120 cm and angled to face the center of the chinrest.A central target speaker was aligned with the mid-sagittal plane directly in front of the chinrest.The six remaining loudspeakers were positioned 20 • , 40 • and 60 • away from the target speaker in both the clockwise and counterclockwise directions.The perturbed location was represented by the "perturbed speaker" placed 40 • away from the target speaker.The adjustable base of the chinrest was positioned at a minimum height of 113.5 cm, and at a distance of 139 cm from the target speaker (so that the participant's ear canal was positioned roughly at the same height as the speakers, at roughly 1.5 m of distance).The average ear canal height for the group was 126.0 ± 2.9 cm.Loudspeakers were obscured from view with a curtain so participants never saw loudspeaker locations.
Each participant was blindfolded with a disposable surgical mask and instructed to keep their eyes closed throughout the experiment.A custom-built sound module shield incorporating an Arduino processor (Arduino Uno/ATmega328P) initiated the presentation of auditory targets and feedback with ~ 100 ms of presentation latency.Two distinct auditory stimuli were amplified using a custom-built 8-channel amplification box with Class D amplifiers (PAM8302A).The mean intensity of 10 iterations of the "target sound" presented from the "target speaker" was recorded at the estimated position of the mean participant's right ear (mean = 46.6 dBA±1.18)using a Digitech sound level meter (QM1592 SPL Meter & Calibrator).Target sound duration was 500 ms and consisted of a 200 Hz series of biphasic pulses, each 90 microseconds in width.The mean intensity of 10 iterations of the "endpoint feedback sound" was recorded in the same manner as the target sound (mean = 59.72 dBA±0.41).The endpoint feedback sound was a broadband white noise stimulus presented for 100 ms with 5 ms rise and fall ramps.Both the target sound and the feedback sound were comprised of a wide bandwidth to support localization.

Headphones group
For the experimental headphones group, the same auditory stimuli used for the loudspeaker condition were filtered through Head-Related Transfer Functions (HRTFs) and presented through wireless headphones (Sennheiser HD 350BT SEBT3) with ~ 140 ms of presentation latency.The mean intensity of 10 iterations of the target sound was recorded at 2 cm from the center of the right headphone speaker (51.77 dBA±0.28),as was the mean intensity of 10 iterations of the endpoint feedback sound (65.35 dBA±0.87).Auditory stimuli were manipulated using HRTFs to convolve the spatial localization characteristics of auditory stimuli presented 1.5 m from the participant.The HRTFs provided interaural timing and volume cues as well as monaural spectral cues associated with a sound's location, to present a headphone simulation of sounds distributed in extrinsic space.HRTF recordings were made in an anechoic chamber from microphones placed in the ears of a participant exposed to sounds at a distance of 1.5 m.Such virtual auditory localization cues have been shown to be as effective as their free-field equivalents (Martin and McAnally, 2001).Similar to the loudspeaker group, participants wore a disposable surgical mask as a blindfold and were instructed to keep their eyes closed throughout the entire experiment (including breaks).

Visual group
For the experimental visual group, participants viewed a monitor (BENQ XL2720-B) set parallel to the floor.The monitor rested on scaffolding that housed the digitizing tablet and obscured vision of the hand.A visual starting-position circle 0.5 cm in diameter was presented above the starting position cradle, which was placed under the monitor's midline, 10 cm from the bottom of the display.When the digitizing pen was placed in the starting position, a transient blue target-circle 0.5 cm in diameter appeared 9 cm directly in front of the starting position for a duration of 500 ms.Participants were instructed to quickly reach and slice through the target in a ballistic straight-line motion.With the exception of the same "too slow" warning used for the loudspeaker and headphones conditions, no sounds were presented in association with task performance.During the baseline block, a circular white cursor 0.5 cm in diameter was superimposed for 50 ms over the target location, informing participants that a reach had been completed successfully.During the perturbation block, this non-veridical endpoint feedback was instead rotated 40 • from the target in either the clockwise or counterclockwise direction (counterbalanced pseudo randomly).At no point during a reach did participants see a veridical cursor representing their actual hand movement.

Motion capture
To restrict head movements and minimize head yaw rotation in particular, head orientations were continuously recorded using reflective markers, infrared cameras (OptiTrack V120: Trio), and Motive software (version 1.8).Baseline head positions directly faced the target stimulus location at a self-reported comfortable chin rest height.Once comfortable, participants were instructed to sit still for 2 s so "baseline head position" could be calculated.For the loudspeaker and visual conditions, three reflective markers were attached to a headband worn by participants.For the headphones condition, three reflective markers were attached to the wireless headphones.Changes in degrees of yaw rotation from baseline head position were recorded during stimulus presentation every trial and monitored online by the experimenter.Following a trial, if head position yaw was rotated more than 5 • from baseline, the experimenter would guide participants back to their baseline head position.Mean head yaw position was slightly clockwise of baseline during the experiment for the loudspeaker group (Mean = 0.43 • ± 1.98 • ), the headphones group (Mean = 0.52 • ± 1.68 • ), and visual group (Mean = 0.56 • ± 1.37 • ).

Data analysis
As participants were exposed either to a clockwise or a counterclockwise error-clamp perturbation, reach-angles from participants experiencing counterclockwise perturbations were reversed for the adaptation and no-feedback blocks, such that positive angles reflected gains in adaptation magnitude for all participants.Prior to data collection, it was decided that data from participants who could not consistently or accurately reach towards the visual or auditory baseline target, indicated by reach-angle standard deviations greater than 20 • , would be removed from analysis.After collecting data from 72 participants, one participant was excluded from the loudspeaker group, three participants were excluded from the headphones group, and no participants were excluded from the visual group.Thus, after collecting data from 4 additional participants, a total of 76 participants were recruited, with 4 data sets being omitted from analysis.Following data collection, each experimental condition contained datasets from 24 participants.
Familiarization.During familiarization, participants made 10 reaches to each target location in random order.We considered two dependent variables to assess reaching performance during familiarization.(1) Signed accuracy was estimated by calculating the average reach-direction deviation from the ideal reach direction for each target.
(2) Variability was estimated by calculating the standard deviation of reach-angles for each target.For both variables, Target (− 60 x Group (Loudspeaker, Headphones, Visual) ANOVAs were run.
Baseline & adaptation.Only 1 straight ahead target was used for all blocks after familiarization.Because reach directions were variable across trials in this experiment, adaptation performance was estimated by mean-averaging reach angles from 40 trials taken from relevant phases of the study.Early learning: To quantify early learning, a Group (Loudspeaker, Headphones, Visual) × Block (Baseline, Perturbation) mixed-design ANOVA was used to quantify the change in reach-angle from the baseline to the first 40 trials of the perturbation block.Adaptation extent.To quantify adaptation extent, a Group (Loudspeaker, Headphones, Visual) × Block (Baseline, Perturbation) mixed-design ANOVA was used to compare reach directions in the last 40 trials of the baseline block with the last 40 trials of the perturbation block.
No-feedback.Sensorimotor adaptation has been shown to decay rapidly upon the removal of feedback (Kitago et al., 2013).Thus, in order to quantify this transient effect, the mean of the first 10 trials of the no-feedback block were used to asses evidence of sensorimotor remapping.To quantify remapping, we ran a Group (Loudspeaker, Headphones, Visual) × Block (Baseline, No-Feedback) mixed-design ANOVA comparing the last 40 trials of the baseline black, to the first 10 trials of the no-feedback block.
An alpha value of 0.05 was applied for all contrasts and Greenhouse-Geisser corrections for violations of the assumption of sphericity were made where necessary.For all t-tests, as we had three groups, we used Fisher's Least Significant Differences procedure, which has been shown to be an appropriate means of performing pairwise comparisons at the established alpha value (Meier, 2006).Figures and ANOVAs were formulated using JASP (Version 0.16.3)statistical software.
Upon conclusion of the experimental block, participants were instructed to continue reaching directly towards the target even though they would no longer be provided with feedback.In the subsequent nofeedback block, movements that remain persistently adapted suggest the presence of implicit remapping.Following the experiment, participants were explicitly asked if they had ever deliberately reached away from a target during the experiment.No participants reported this to be the case.

Familiarization
Fig. 2 shows raw familiarization reaching trajectories for example participants from each group.
We first quantified differences in reach accuracy to multiple auditory and visual targets during the familiarization block, in order to compare sensorimotor reach accuracy across modalities.Signed reach-angle error differed significantly across targets (main effect of Target-direction, F (2.27, 156.36.74) = 58.552,p < 0.001, partial-eta-squared = 0.459; Fig. 3A), as indicated by positive and negative biases for different targets.The main effect of Group was not significant (F (2, 69) = 2.566, p = 0.084, partial-eta-squared = 0.069), but there was a significant interaction between Target-direction and experimental Group (F(4.53,156.36) = 4.061, p = 0.002, partial eta-squared = 0.105), which showed B. Miller-Mills et al. that target specific clockwise and counter-clockwise biases in reaches to targets presented by headphones were greater than those presented by loudspeakers and by visual targets.
Reach performance to the central 0 • target was of particular interest, as this was the target location used for the remainder of the experiment.Thus, we examined the standard deviation of the central target across all experimental conditions.Participants demonstrated the largest mean standard deviation of reach-angle to the 0 • target for the headphones condition (16.1 • ± 10.6 • ) followed by the loudspeaker condition (9.9 • ± 6.5 • ), and then the visual condition (7 • ± 2.9 • ).Mean standard deviations were significantly different between the visual group and headphones group (t(23) = -3.78,p < 0.001), as well as the loudspeaker group and headphones group (t(23) = -2.17,p = 0.041), but not significantly different between the visual group and loudspeaker group (t(23) = -1.93,p = 0.066).

Experimental movement times
We gave participants feedback to encourage them to reach quickly with an uncorrected shooting motion.To check that this was effected, we report the group movement times for the 480 trials that follow the familiarization block in Fig. 5 (visual group = 0.058 ms ± 0.042 ms,  loudspeaker group = 0.163 ms ± 0.031 ms, headphones group = 0.161 ms ± 0.026 ms; main effect of group, F (2, 69) = 76.79,p < 0.001).Mean movement times were significantly shorter for the visual group than the loudspeakers group (t(23) = 8.90, p < 0.001), and shorter for the visual group than the headphones group (t(23) = 10.73,p < 0.001).There was no significant difference between the loudspeaker group and headphones group (t(23) = 0.40, p = 0.696).Thus, although all participants in the three groups had shorter movement times than our criterion (0.3 s) and performed fast shooting movements, the participants in the visual group reached considerably faster than those in the auditory groups, despite also triggering the most "too slow" warnings.

Adaptation
Fig. 6 shows baseline subtracted reach angles for the three groups throughout the experiment.Positive angles indicate a difference from baseline in the direction required to compensate for the error-clamp perturbation.All groups showed a corrective response upon exposure to the perturbation, and a gradual return to baseline during the washout phase, but the time-course and magnitude of the adaptive effects differed between groups, particularly for early adaptation (Fig. 7).

Aftereffect
For the final no-feedback block, participants were explicitly instructed that the perturbation had been removed, and they received no auditory or visual feedback while reaching.Adapted reaches in such trials are typically interpreted as evidence of implicit remapping.Implicit remapping was evidenced by a significant main effect of Block (F (1, 69) = 15.895,p < 0.001, partial-eta-squared = 0.187), with no significant effect of Group (F (2, 68) = 0.738, p = 0.482).The magnitude of implicit adaptation differed for the three conditions, as supported by a significant Group x Block interaction (F(2, 69) = 4.252, p = 0.018,

Relationship between reaching speed, baseline standard deviation, and adaptation magnitude
The observed differences in movement time shows that people performed the reaching task differently with and without vision on average, so we checked if there was a hint that adaptation to the error-clamp was influenced by the speed of reaching movement.We found no significant correlation between average movement times and adaptation magnitude for any of the three groups (all R 2 < 0.15).We also observed standard deviation differences between groups when reaching to the baseline target (Fig. 4b).We found no significant correlation between adaptation magnitude and the standard deviation of baseline reach direction for any of the three groups (all R 2 < 0.15).We therefore consider it unlikely that differences in movement speed, or the standard deviation of baseline reach direction, were important factors in distinguishing adaptation magnitude between groups.

Discussion
This study demonstrates that reaching movements can be adapted to correct for imposed feedback errors from the visual and auditory modalities.Because participants were instructed to ignore the perturbed feedback during the error-clamp block, and because reach directions remained significantly adapted after participants were informed that the perturbation was removed, we interpret these results as evidence of implicit visuomotor and audiomotor adaptation.Although reach directions were generally more variable and less accurate to auditory targets than visual targets, our audiomotor error-clamp paradigm was sufficient to generate measurable adaptation for two different presentations of auditory feedback (headphones and loudspeakers).Implicit sensorimotor adaptation was measurable despite task features that are sub-optimal for driving the adaptive process, including movements to remembered locations, a lack of online movement feedback, and a brief duration of endpoint feedback.Considering these constraints, the mechanisms responsible for implicit sensorimotor adaptation appear robust and ubiquitous across individuals.Alternatively, it remains possible that novel task features from the present study may have resulted in a different form of implicit adaptation relative to standard visuomotor paradigms.However, we are unaware of an alternative adaptive mechanism that has been documented to describe these effects.
The capacity to elicit implicit audiomotor adaptation offers an opportunity to address general questions regarding sensorimotor remapping, by expanding a range of sensorimotor learning paradigms to the audiomotor domain.In particular, future work that characterizes differences between implicit visuomotor and audiomotor adaptation may provide further insight into how sensory to motor transformation are modified following a sensory prediction error, as well as possible crossmodal effects.
Reach direction adaptation has been previously reported in response to a directionally informative auditory error-tone, played continuously throughout a reach (Schmitz and Bock, 2014).With this approach, the deviation of hand direction from a straight line to the target was Fig. 6.Data points represent the average reach-angle of 1 reach for all 24 participants in each group.Each reach was directed to the same 0 • target used during the Familiarization Block.Shading indicates the standard error across each group.The mean of the last 80 trials of the Baseline block has been subtracted from each data point to emphasize the changes in reach-angles from baseline to the perturbation and no-feedback blocks.
signalled by a change in the frequency of the tone, and this allowed participants to make online corrections to reach directions.This somewhat abstract mapping between hand direction and auditory stimuli differs substantially from the form of auditory feedback used in the current study, in which location of auditory feedback was varied.For the present study, as the auditory feedback was triggered at the end of a reach, we consider this form of endpoint feedback to have some ecological validity, as it parallels the spatiotemporal features of an extended hand making contact with a physical object, a process especially important in low light conditions.However, as continuous online visual feedback typically increases sensorimotor adaptation compared to error-clamp perturbations that offer only endpoint feedback (Tsay et al., 2021a;Wang et al., 2024), future work should investigate the impact of online feedback during implicit audiomotor adaptation.
The magnitude of adaptation during the final 40 error-clamp trials was small for the loudspeaker (~5 • ) and visual groups (~3 • ), and larger for the headphones group (~17 • ).Aftereffects were also correspondingly small (~3 • , ~ 3 • , and ~ 13 • respectively).We interpret adaptation estimates for the headphones group with caution, as there were large biases and considerable variability in reach performance for this condition.Ultimately, we observed audiomotor adaptation magnitudes that were less than the ~ 20 • of reach-angle change reported in response to auditory stimuli by Schmitz and Bock (2014).Several experimental design differences may explain this: including the fundamentally different nature of the feedback, the number of targets locations presented, the size of the perturbation, and the fact that participants had their eyes open in the study by Schmitz and Bock.However, most importantly, because it is unclear whether Schmitz and Bock informed participants that the perturbation had been removed when they measured aftereffects, it is possible that the reported adaptation reflected explicit re-aiming, rather than implicit motor learning.In contrast, participants for our experiment were always instructed to reach directly towards the target stimulus, there was no task-related reason to change this behavior, and participants were explicitly told to ignore the perturbation from its first trial.Additionally, instructions were given that the perturbation had ended prior to the final nofeedback block.We therefore conclude that the change in reach-angle observed during the no-feedback block reflects an implicit remapping of sensory information.
The observed adaptation extent for our visual group was also smaller than reported in previous error-clamp experiments, where adaptation magnitudes between ~ 10-15 • have typically been reported (Morehead et al., 2017;Poh and Taylor, 2019;Vandevoorde and Orban de Xivry, 2019).We suspect this is due to the incorporation of task features chosen to match the auditory tasks, in particular, the brief target and endpoint feedback durations.The time-course of adaptation for our visual group was rapid (0.2 • for perturbation trial 1, 2.7 • for perturbation trial 2, 2.9 We suspect the common notion that implicit adaptation is slow to develop relates to studies highlighting visuomotor rotation responses as a combination of explicit and implicit re-aiming (e.g.Mazzoni & Krakauer, 2006;Taylor, Krakauer, & Ivry, 2014).In these studies, explicit adaptation is shown to be strikingly rapid and flexiblesuch that entire 45 • perturbations can be corrected within a couple of trials with no aftereffect.By contrast, implicit adaptation is clearly more gradual, progressive, and persistent, and it elicits a longer lasting and measurable aftereffect.However, as further study has brought to light, this framework may be more nuanced.Ruttle and colleagues have documented To address differences in adaptation between the visual and auditory sensory modalities, several psychophysical processes must be considered.Adaptation magnitudes may vary due to differences in sensory acuity for localization, differences in how each modality updates the motor plan following a sensory prediction error, or differences in the way that egocentric space is referenced by each modality.It therefore remains unclear if spatial representations typically driven by visual processes contribute to the development and expression of implicit audiomotor adaptation.Thus, as participants never received feedback from both visual and auditory modalities during this experiment, crossmodal designs will be needed to further probe the relationship between visual and auditory adaptation.
Harris & Craske demonstrated that when participants wore prismglasses and achieved their limit of visuomotor adaptation, they showed essentially the same degree of error when reaching with eyes closed to auditory targets, as with their eyes open towards visual targets (Craske, 1966).The precise contribution of the implicit motor learning system during such a task is unclear, however, as this study did not focus on isolating implicit learning.Furthermore, although audiomotor aftereffects resulting from visual perturbations are relevant for crossmodal generalization, the study by Harris & Craske did not assess whether audiomotor aftereffects can occur as a consequence of entirely auditory perturbations.
In addition to demonstrating implicit adaptation to clamped auditory feedback, our experiment may also provide insight into the features of auditory-guided reaching behavior.For both auditory conditions, the familiarization target located on the interaural midline (i.e. the 0 • target) demonstrated the greatest reach-angle standard deviation during the initial familiarization block.This is not surprising considering the mechanisms responsible for auditory localization.For humans, auditory localization is driven by the time it takes for a sound to reach each ear (interaural timing differences), as well as the difference in intensity between ears when processing a sound (interaural level differences) (Feddersen et al., 1957).Thus, as supported by our results, target location information becomes less ambiguous when auditory stimuli contain asymmetrical binaural cues (cues not located on the interaural midline).This likely explains our finding that, for both the loudspeaker and headphones groups, participants were less variable in reach direction towards targets not located on the interaural midline.However, as interaural differences were designed to be similar between the loudspeaker and headphones groups, alternative explanations must be considered in order to account for the substantial difference in adaptation magnitude between the two conditions.
Perhaps the most impactful methodological difference between the auditory conditions was the unobstructed outer-ear and pinnae of participants exposed to loudspeakers.It has been shown that obscuring a participant's pinna affects their ability to localize auditory cues presented within the same quadrant of azimuth along the horizontal plane (Musicant and Butler, 1984).Thus, while interaural differences drive auditory detection, small features of the pinnae that vary across individuals also provide information about the location of sounds via unique monaural spectral cuesa process significantly contributing to azimuthal localization.Without these individualized monaural spectral cues, participants commonly mislocalize broadband auditory targets by up to 30 • within the same quadrant of azimuth prior to front/back reversal corrections (Musicant and Butler, 1984).Differences in reach accuracy and adaptation between the auditory groups may therefore have been driven by a lack of individualized monaural spectral cues for the headphones participants.Effectively, where individuals were using their own head dimensions and pinnae to assign customized monaural spectral cues in the loudspeaker group, monaural spectral cues were not customized for the headphones group, likely providing an additional measure of noise and uncertainty.
Perturbations of auditory localization may also aid in future studies that probe cross-modal implicit motor learning generalizability.As such, our results offer a foundation for future study involving new clinical populations, as well as for work examining the sensorimotor adaptation of clinical populations previously probed visually.Specifically, research involving people with congenital vision impairment may reveal the degree to which sensorimotor adaptation relies on a foundation of visual localization mechanisms.Likewise, research with populations who have cerebellar ataxia may inform whether audiomotor adaptation relies on the same cerebellar-dependent mechanisms as visuomotor adaptation.The current data suggest that systematic reach direction errors lead to implicit reach direction adaptation, irrespective of whether the errors are signaled by visual or auditory feedback.This raises the possibility that there might be an obligatory neural process to recalibrate sensory to motor maps that is ubiquitous across sensory modalities.Future work should determine whether such recalibrations change characteristics during unimodal versus multimodal sensorimotor transformations.

Fig. 1 .
Fig. 1.The Left Panel depicts the testing apparatus: including a chin rest, a tablet obscured from view (containing a U-shaped plastic cradle positioned around the hand's starting position), and a horizontally positioned monitor.The Right Panel depicts locations of the seven loudspeaker familiarization targets.

Fig. 2 .
Fig. 2. Raw familiarization reaching trajectories for example participants from each group.

Fig. 3 .
Fig.3.Reach-angle errors to each of the seven familiarization block targets.The target listed at 0 • was positioned directly in front of the participant's mid-sagittal plane, and was the target location used during all subsequent blocks.Negative target locations were counterclockwise of the 0 • target.Fig.3Ashows group-means for hand angle error, with positive reach-angle errors indicating bias to the clockwise direction.Fig.3Bshows group means for reach direction standard deviations to each of the seven familiarization targets.

Fig. 4 .
Fig. 4. Group averages for each group's final 40 baseline trials.Error bars represent the dataset range.A box's top boundary indicates the upper quartile, a box's bottom boundary indicates the lower quartile, and the bisecting horizontal line indicates the dataset median.The 0 • target location was used each trial.Fig. 4A shows reach-angle error group means.Degrees of positive reach-angle error indicate inaccuracy in the clockwise direction.Fig. 4B shows group standard deviations.

Fig. 5 .
Fig. 5.For each group, movement times are the average of all experimental trials following the familiarization block.Error bars represent the dataset range.A box's top boundary indicates the upper quartile, a box's bottom boundary indicates the lower quartile, and the bisecting horizontal line indicates the dataset median.
• for perturbation trial 3), yet in line with other error-clamp studies showing individual trials post-perturbation (Poh and Taylor, 2019: ~6• for perturbation trial 1; Tsay et al 2021b: ~1• for perturbation trial 1), as well as the original per trial early adaptation estimates made by Morehead et al. (2017) in response to a 45 • error-clamp (~1 • per trial).

Fig. 7 .
Fig. 7. Fig. 7A shows the difference between the mean of the final 40 Baseline reaches, and the mean of the first 40 perturbation reaches.Error bars represent the standard error.For Fig. 7B, 7C and 7D, Baseline data points represent the mean of the final 40 reaches.Adapt data points represent the mean of the first 40 perturbation reaches.Y-axis scales differ between panels in order to visualize performance variability across participant samples.

Fig. 8 .
Fig. 8. Fig. 8A shows the difference between the mean of the final 40 Baseline reaches, and the mean of the final 40 perturbation reaches.Error bars represent the standard error.For Fig. 8B, 8C and 8D, Baseline data points represent the mean of the final 40 baseline reaches.Adapt data points represent the mean of the final 40 perturbation reaches.Y-axis scales differ between panels in order to visualize performance variability across participant samples.

Fig. 9 .
Fig. 9. Fig. 9A shows the difference between the mean of the final 40 Baseline reaches, and the mean of the first 10 no-feedback reaches.Error bars represent the standard error.For Fig. 9B, 9C and 9D, Baseline data points represent the mean of the final 40 baseline reaches.AE data points represent the mean of the first 10 nofeedback reaches for a participant.Y-axis scales differ between panels in order to visualize performance variability across participant samples.