A locomotor learning paradigm using distorted visual feedback elicits strategic learning

Abstract

Distorted visual feedback (DVF) during locomotion has been suggested to result in the development of a new walking pattern in healthy individuals through implicit learning processes. Recent work in upper extremity visuomotor rotation paradigms suggest that these paradigms involve implicit and explicit learning. Additionally, in upper extremity visuomotor paradigms, the verbal cues provided appear to impact how a behavior is learned and when this learned behavior is used. Here, in two experiments in neurologically intact individuals, we tested how verbal instruction impacts learning a new locomotor pattern on a treadmill through DVF, the transfer of that pattern to overground walking, and what types of learning occur (i.e., implicit vs. explicit learning). In experiment 1, we found that the instructions provided impacted the amount learned through DVF, but not the size of the aftereffects or the amount of the pattern transferred to overground walking. Additionally, the aftereffects observed were significantly different from the baseline walking pattern, but smaller than the behavior changes observed during learning, which is uncharacteristic of implicit sensorimotor adaptation. Thus, experiment 2 aimed to determine the cause of these discrepancies. In this experiment, when VF was not provided, individuals continued using the learned walking pattern when instructed to do so and returned toward their baseline pattern when instructed to do so. Based on these results, we conclude that DVF during locomotion results in a large portion of explicit learning and a small portion of implicit learning.

NEW & NOTEWORTHY The results of this study suggest that distorted visual feedback during locomotor learning involves the development of an explicit strategy with only a small component of implicit learning. This is important because previous studies using distorted visual feedback have suggested that locomotor learning relies primarily on implicit learning. This paradigm, therefore, provides a new way to examine a different form of learning in locomotion.

INTRODUCTION

Historically locomotion has been thought of as a predominantly implicit task; however, community ambulation and changes to gait during rehabilitation are thought to involve more conscious control. Despite this, locomotor learning has been examined using primarily implicit paradigms, specifically the split belt treadmill (Choi and Bastian 2007; Reisman et al. 2005, 2010). The split belt treadmill paradigm, in which two independent treadmill belts drive the legs into asymmetry by forcing stepping at two different speeds, is thought to drive sensorimotor adaptation, an implicit learning process in which a movement is changed in response to a sensory prediction error (Martin et al. 1996). In the upper extremity, visuomotor rotation paradigms have been used to understand sensorimotor adaptation (Krakauer et al. 2000). With visuomotor rotation paradigms as a framework, several locomotor studies have used a visuomotor walking task with distorted visual feedback (DVF) to drive adaptation of step length during locomotion (Kim and Krebs 2012; Kim and Mugisha 2014; Kim et al. 2015). Regardless of awareness of the distortion, distraction, or instructions to ignore the visual feedback (VF), subjects learned an asymmetric walking pattern (Kim and Krebs 2012; Kim and Mugisha 2014; Kim et al. 2015). Based on these findings, the authors suggested that the learning occurring during the DVF locomotor paradigm was a purely implicit process resulting from sensorimotor adaptation (Kim and Krebs 2012; Kim and Mugisha 2014; Kim et al. 2015).

These studies provide insight into the ability of neurologically intact adults to learn a new walking pattern with DVF but do not provide insight into how well the walking pattern learned though DVF transfers from one context to another (e.g., the treadmill to overground walking). In upper extremity lifting, it has been found that the information provided before the learning task impacted the transfer of a newly learned lifting technique (Fercho and Baugh 2016). More specifically, Fercho and Baugh found that healthy individuals who were told that the errors they experienced during learning were caused by the environment did not transfer the newly learned lifting technique as well as individuals who were not told that the environment would cause errors (Fercho and Baugh 2016). It is unclear if the same would be true in locomotion. As a result, in experiment 1, we aimed to determine how the information provided about the DVF before learning a new asymmetric walking pattern would impact acquiring the new pattern and then transferring it from treadmill to overground walking. Based on the previous work suggesting that learning a new walking pattern through DVF relies heavily on implicit learning (Kim and Krebs 2012; Kim and Mugisha 2014; Kim et al. 2015), we hypothesized that individuals who were aware of the distortion of the VF would learn the pattern to a greater extent, but demonstrate less transfer of the new pattern, than those who were unaware of the distortion.

Because of the results of experiment 1, we began to question the previous assertion that learning during the locomotor DVF paradigm occurred through implicit sensorimotor adaptation (Kim and Krebs 2012; Kim and Mugisha 2014; Kim et al. 2015). Hussain et al. (2013) found that neurologically intact individuals were able to change their swing time in response to DVF; however, when instructed to recall the learned pattern 24 h later, subjects were able to do so (Hussain et al. 2013), suggesting that there may be an explicit component in learning through DVF. Additionally, Kim et al. (2015) found that retention of the learned pattern was impacted by the information provided before learning. Specifically, individuals who were naïve to the DVF continued to produce the asymmetric walking pattern when the VF was removed while the group informed about the DVF returned to a more symmetric walking pattern. Although the authors of that study did not conclude that there is an explicit component of the DVF paradigm, their result suggests that there may be both implicit and explicit learning processes occurring with this paradigm.

This is supported by recent work in the upper extremity that suggests that visuomotor rotation paradigms in reaching result in both implicit and explicit learning processes (Haith and Krakauer 2013; Mazzoni and Krakauer 2006; Taylor and Ivry 2014). In visuomotor rotation tasks, the explicit portion of this learning is thought to represent the development of a strategy, while the implicit portion has been shown to result from sensorimotor adaptation (Haith and Krakauer 2013; Mazzoni and Krakauer 2006; Taylor et al. 2014). Unlike implicit sensorimotor adaptation, strategic learning is thought to be driven by task performance errors (i.e., the successful completion of the task’s goal) (Taylor and Ivry 2014). Past work with healthy individuals suggests that individuals can be prompted to learn through strategic methods with verbal and environmental cues (Benson et al. 2011; Mazzoni and Krakauer 2006; Morehead et al. 2015). Thus, in experiment 2, we aimed to understand the types of learning (i.e., implicit or explicit) that occurred with the DVF locomotor paradigm. We hypothesized that the DVF locomotor paradigm was dominated by a form of learning similar to the explicit strategic learning process, which would result in the selection of a different walking pattern based on the verbal instruction provided.

METHODS

Subjects

Subjects were recruited from the local community using online advertisements and emails to listservs within the University of Delaware community. Individuals between the ages of 18 and 40 with no history of neurological conditions (including attention deficit hyperactivity disorder and concussion) or pain that limited walking were included. Individuals were excluded if they had had a cardiac event within the past 3 mo, unexplained dizziness within the past 6 mo, or visual deficits that were not corrected. All participants signed a written informed consent approved by the Human Subjects Review Board at the University of Delaware before participation.

Experimental Set Up

During treadmill walking in both experiment 1 and 2, subjects walked with a 20-inch computer monitor positioned 50 inches in front of the treadmill to display real-time VF regarding the subject’s step length (SL) (Fig. 1A). Both experiments consisted of one session with four phases: Baseline, Orientation, Learning, and Postlearning (Fig. 1B). During Orientation and Learning in both experiments 1 and 2, subjects walked on the treadmill while the real-time display was created using The Motion Monitor Toolbox (Innovated Sports Training, Chicago, IL). This feedback was in the form of a bar graph, with one vertical bar representing the right leg and one for the left leg (Fig. 1B). During the swing phase of a leg, the bar corresponding to that leg grew in height until heel strike was detected. Once heel strike was detected, the bar maintained that size until the next swing phase on that leg was initiated. Heel strike was identified as the moment the vertical position of the heel marker was below a specific threshold (defined by the marker placement in standing) and ground reaction force exceeded 20 Newtons. Additionally, a target line was positioned two-thirds of the way up the screen, which represented the subject’s baseline SL on the treadmill (Fig. 1B). When the bar representing SL was stopped at heel strike within 1.5% (plus or minus) of the target line, the words “great” or “awesome” appeared on the screen to provide positive feedback.

Fig. 1.

Experimental set up. A: during all treadmill walking, a computer monitor was positioned in front of the subject. Visual feedback (VF) was displayed during Orientation and Learning phases. B: subjects in both experiments participated in the following four phases: Baseline, Orientation, Learning, and Postlearning. During Orientation and Learning, VF about step length (SL) was displayed. The display consisted of two bars, one for each leg, and a horizontal target line representing their baseline SL. As each leg initiated its swing phase, the bar representing that leg grew until heel strike occurred. During Learning, the VF was distorted so that the bar representing the left leg was shorter than the actual SL (represented by the broken line) while the bar representing the right leg was longer than the actual SL (represented by the broken line).

Orientation was conducted with accurate VF to provide the subjects with exposure to the display before Learning and ensure that the subjects understood the display. Before Orientation subjects were told, “You are going to see a bar graph on the screen in front of you. Each bar will represent your step length. Your step length is the distance from one heel to the other when the front heel hits the ground. For the first minute, you can change your step length to see how it works. After a minute, I will ask you to return to comfortable walking.” The researcher also demonstrated SL while providing these instructions. To ensure subject understanding, the subject verbally described and demonstrated SL back to the researcher. Additionally, subjects were asked to volitionally change their SL to ensure they understood the display. The data from this phase were not included in our analysis, since it was solely for the purpose of introducing the subjects to the display.

During Learning, the VF on the display was distorted as a percentage of SL during Baseline. The amount of distortion was equal, but opposite, for each leg. For all subjects, the left bar was distorted to make it appear that they took a shorter step than they actually did while the right bar made it appear that they took a longer step than they actually did (Fig. 1B); thus, to have the bars match the target during distortion, the subject needed to take a longer step with their left leg and a shorter step with their right leg. For both experiments, the DVF was introduced gradually over the course of 8 min. During the 1st min, the left leg display was 1% of baseline shorter than their actual SL, whereas the right leg display was 1% of baseline longer than their actual SL. Each minute the distortion percentage increased by 1% until the beginning of the 9th min when each individual leg was distorted by 9%. This amount of distortion was constant for the remaining 2 min of Learning (Figs. 2A and 3A).

Fig. 2.

Experiment 1 paradigm and results. A: experimental paradigm for experiment 1. B: stride-by-stride data in bins of 3 for the Instruction (INS, red) and No Instruction (No INS, blue) groups. Shaded region indicates SD across subjects.

Fig. 3.

Experiment 2 paradigm and results. A: experimental paradigm for experiment 2. B: stride-by-stride data in bins of 3 for the INS Post group. Shaded regions indicate SD.

Experimental Protocol

Experiment 1.

Subjects were randomized into two groups: Instruction (INS) or No Instruction (No INS). All subjects participated in overground and treadmill walking (Fig. 2A). All overground walking was performed at the subject’s self-selected walking speed and consisted of 10 trials down a 5-meter walkway. All treadmill walking was performed at 1.2 m/s on a Bertec (Columbus, OH) treadmill. Baseline consisted of overground and treadmill walking (Fig. 2A). During the treadmill portion of Baseline, the subject’s SL was calculated to establish the target line and a reference for the distortion as described above. Following this, all subjects participated in Orientation as described above. Subjects then walked for 15 min during Learning. This time was broken into two sections; the first was 10 min, and the second was 5 min (Fig. 2A). Between these two sections was a 30-s Catch Trial with no VF. The 10-min section of Learning was before the Catch Trial, and the VF was gradually distorted as described above. The remaining 5 min of Learning were after the Catch Trial. During this last portion of Learning, the level of distortion was constant at 9% for each leg (i.e., the maximum level achieved before the Catch Trial). Before Learning, the INS group was told, “You will now see the same bars that you saw for the last 2 min. The green bar represents your right step length, and the blue bar represents your left step length. You will also see a horizontal line on the screen. I would like you to keep the bars at that line to receive the positive feedback. To do this, you will need to gradually increase or decrease your step length because the feedback may not be accurate all the time.” The No INS group was told, “You will see the same display the whole time.” Before the Catch Trial, subjects were told that they would not see the display and to simply walk. At the completion of Learning, subjects participated in Postlearning without VF, which consisted of overground walking in the same manner as Baseline. During this phase, subjects were instructed to simply walk.

At the end of walking, a questionnaire was completed to determine if the subjects were aware of the distortion (No INS group) and what specifically they felt had occurred (both groups). As detailed under results, the aftereffects observed, particularly during the Catch Trial, were different than expected; thus, we added experiment 2 to further explore the types of learning occurring with the locomotor DVF paradigm.

Experiment 2.

A new group of subjects (INS Post) attended one session with treadmill walking at 1.2 m/s (Fig. 3A). The INS Post group first performed Baseline and Orientation phases as in experiment 1, followed by a 10-min Learning phase. Before Learning, all subjects were provided with the same instructions as the INS group from experiment 1. At the end of 9 min, the INS Post group was told, “Try and remember exactly what you are doing to get the bars to match the pink (target) line. In about a minute, I am going to stop the treadmill, and later in the study I will ask you to use the same walking pattern that you are using now.” These instructions were adapted from Hussain et al. (2013). At the end of Learning, subjects participated in three, 5-min Postlearning phases, all without VF: 1) Retention 1, in which participants were asked to continue walking with the same pattern they had just learned on the treadmill, 2) Normal Walking, in which participants were asked to return to their normal walking pattern, and 3) Retention 2, in which participants were asked to return to the walking pattern that they learned during the Learning phase.

Data Analysis

During all walking tasks, kinematic data were collected at 100 Hz using an eight-camera Vicon Motion Capture system (Vicon MX, Los Angeles, CA). Six retroreflective markers were placed on bilateral heels, lateral malleoli, and fifth metatarsal heads. A custom-written Matlab script (MathWorks, Natick, MA) was used to calculate our primary measure of SL, which was defined as the sagittal distance between right and left heel marker when heel strike occurred (Zeni et al. 2008). Learning variables for both experiments were calculated using a step symmetry index (SSI) as follows:

$$\text{SSI}=100\times \frac{{\text{SL}}_{\text{Left}}-{\text{SL}}_{\text{Right}}}{{\text{SL}}_{\text{Left}}{\text{+SL}}_{\text{Right}}}$$

A SSI of zero indicates perfect symmetry, whereas a positive SSI indicates a longer step with the left leg compared with the right.

Statistical Analysis

All statistical analyses were performed in SPSS. For all measures, normality and homogeneity were assessed using Kolmogorov-Smirnov and Levene’s tests, respectively. In experiment 1, independent t-tests were used to compare SSI during Learning, Catch Trial, and Postlearning between groups. Additionally, a paired t-test was used to compare SSI at Baseline and Catch Trial. If normality was not met, nonparametric tests were used. In experiment 2, paired t-tests were used to compare SSI during 1) Retention 1, Normal, and Retention 2 with Baseline and 2) Retention 1, Normal, and Retention 2 with Learning within the INS Post group (Fig. 3B). This was done to determine if individuals could select the correct pattern (i.e., the learned pattern or the baseline pattern) based on a verbal cue. If normality was not met, nonparametric tests were used. Bonferonni corrections were applied for multiple comparisons.

RESULTS

Experiment 1

Thirty-three subjects were enrolled in this experiment; however, five individuals were excluded before data analysis because of technical difficulty (n = 1), an asymmetric baseline step length >5 cm (n = 1), being unable to adjust step length during the TM orientation phase (n = 1), or not learning an asymmetric pattern of at least 5% (n = 2). Thus, data from 28 individuals (23.4 ± 3.81 yr; 11 males) were used with 14 subjects per group. Based on the results of the questionnaire administered at the end of the study, all but four subjects (3 from No INS and 1 from INS) were able to identify that they were walking differently. In the INS group 12 of the 14 subjects could specifically identify what they were doing differently with at least one of the legs during walking, whereas only 5 of the 14 in the No INS group were able to identify what they were doing differently with at least one leg.

Figure 2B shows the behavior of the INS and No INS groups over the course of the paradigm. At Baseline, subjects were similarly symmetric (i.e., SSI = 0) for both overground [t(26) = −0.10, P = 0.919, d = −0.04] and treadmill [t(26) = −1.30, P = 0.205, d = −0.49] walking. Both groups learned an asymmetric walking pattern during Learning; however, the INS group learned a significantly more asymmetric pattern than the No INS group during the last 30 strides of Learning before the Catch Trial [t(26) = 3.14, P = 0.004, d = 1.19] and the last 30 strides of the Learning post Catch Trial [t(26) = 3.27, P = 0.003, d = 1.23; Fig. 4A]. This difference was observed at the end of Learning before the Catch Trial, immediately after the Catch Trial, and at the end of Learning before Postlearning (Fig. 2B). Despite the difference in SSI between the two groups during Learning, there was no difference between groups during the first 10 strides of the Catch Trial [t(26) = −0.723, P = 0.476, d = −0.27; Fig. 4B]. A paired t-test showed a significant difference between SSI during Baseline treadmill walking and the Catch Trial within each group [No INS: t(13) = −3.92, P = 0.002, d = −1.48; INS: t(13) = −4.57, P = 0.001, d = −1.73], with the walking pattern during the Catch Trial being more asymmetric than at Baseline (Fig. 4B). This indicates that there was an aftereffect following learning, but that the aftereffect was similar between groups. Additionally, there was no difference between groups during transfer [t(26) = 0.027, P = 0.98, d = 0.01] and no difference within groups between the overground Baseline and Postlearning [No INS: t(13) = −2.07, P = 0.059, d = −1.04; INS: t(13) = −1.335, P = 0.205, d = −0.67], suggesting that there was no transfer of the newly learned pattern to overground walking in either group.

Fig. 4.

Amount learned and stored during experiment 1. A: the Instruction (INS) group (gray) learned a significantly more asymmetric walking pattern than the No Instruction (No INS) group (black). This was the case during the last 30 strides of Learning before the Catch Trial (labeled Learning Precatch) and the last 30 strides of Learning (labeled End of Learning). B: both the INS (gray) and No INS (black) groups walked with a significantly more asymmetric pattern during the Catch Trial than at Baseline; however, there was no difference in asymmetry between groups during the Catch Trial. Each dot represents an individual subject, and the error bars represent SD.

The results of the Catch Trial were surprising for two reasons. First, because of the differing levels of asymmetry that the groups achieved during Learning, one would expect the magnitude of the aftereffects to be different between groups. Additionally, the size of the aftereffects observed relative to how much was learned (Fig. 2B) was much smaller than would be expected if this paradigm drove implicit sensorimotor adaptation (Kim and Krebs 2012; Kim and Mugisha 2014; Kim et al. 2015). As a result, we sought to explore the types of learning occurring in this paradigm (i.e., explicit vs. implicit). This line of questioning is supported by upper extremity work demonstrating both implicit and explicit learning processes occurring during visuomotor rotation tasks (Taylor and Ivry 2012; Taylor et al. 2014). The use of the explicit processes, typically thought of as the deployment of a strategy, can be prompted with both verbal and environmental cues (Benson et al. 2011; Hussain et al. 2013; Mazzoni and Krakauer 2006; Morehead et al. 2015). Based on this work, it is possible that subjects in both groups simply selected not to use the strategy learned during the Learning phase during the Catch Trial and Postlearning phases because there was no reason to do so (i.e., no verbal instructions or VF to indicate a need to deploy the learned strategy). To further assess whether this paradigm promoted explicit strategic learning, we performed experiment 2.

Experiment 2

Six new subjects (24.3 ± 7.1 yr; 2 males) were enrolled in the INS Post group. Figure 3B shows the changes in walking pattern of the INS Post group. Subjects learned a significantly more asymmetric walking pattern during the Learning phase (Fig. 3B). The first 10 strides of Retention 1 and Retention 2 were significantly different from Baseline [t(5) = −17.40, P = 0.003, d = −9.64 and t(5) = −3.515, P = 0.016, d = −1.66, respectively]. The walking pattern used during Retention 1 and 2 was not significantly different from the last 30 strides of Learning [t(5) = −0.439, P = 0.679, d = −0.23; t(5) = 0.421, P = 0.691, d = 0.22, respectively]. Next, there was a significant difference between the last 30 strides of Learning and the first 10 strides of Normal Walking [t(5) = 12.991, P < 0.001, d = −0.80; Fig. 5A]. Last, comparing the first 10 strides of the Normal Walking with Baseline, we found no significant difference [t(5) = −2.37, P = 0.063, d = −0.61; Fig. 5B], although, not significantly different from Baseline, there appeared to be a trend toward increased asymmetry despite being asked to walk normally. This is further supported by the fact that there was not a significant difference between the first 10 strides of Normal Walking in INS Post and the Catch Trial in the INS group in experiment 1 [t(18) = −0.38, P = 0.71, d = −0.19; Fig. 5B]. Taken together, the results from the INS Post group show that subjects were able to continue using a newly learned asymmetric walking pattern when asked to do so but were also able to switch back toward their normal walking pattern when instructed to do so, but still with a small aftereffect similar to that seen in experiment 1 during the Catch Trial.

Fig. 5.

Learning and Postlearning during experiment 2. A: when individuals in INS Post were asked to continue using the learned pattern (Retention 1 and 2), there was no significant difference in step symmetry index (SSI); however, when asked to return to their normal walking pattern (Normal Walking) subjects walked with a significantly more symmetric pattern than at the end of Learning. B: there was a significant difference between SSI during Baseline and the Catch Trial in the INS group (gray) and a trending difference in INS Post (black). To support this trend, there was no significant difference between INS and INS Post during Normal Walking and the Catch Trial. Each dot represents an individual subject, and the error bars represent SD.

DISCUSSION

Impact of Awareness of Distortion on Learning and Transfer

Based on the results of experiment 1, awareness of the distortion in the DVF locomotor paradigm resulted in more complete learning of the task compared with subjects who were unaware of the distortion. However, awareness of the distortion did not impact transfer of the newly learned pattern to overground walking. This is likely because of the magnitude of the aftereffects observed during the Catch Trial, which were quite small relative to the variability of overground walking.

The impact of awareness on the amount learned is supported by work in the upper extremity, which found that awareness of the perturbation resulted in more complete learning of an upper extremity visuomotor rotation task (Neville and Cressman 2018). Additionally, Roemmich et al. (2016) found that, when subjects were provided with visual feedback and instruction to keep their step length symmetric during split belt treadmill walking, subjects had a faster rate and greater magnitude of learning. The authors suggest that information added through visual feedback resulted in sensorimotor adaptation and an additional more explicit form of learning that they refer to as “voluntary correction” (Roemmich et al. 2016). Based on the results of experiment 1 and previous studies, we believe that there are two learning processes occurring with the present DVF locomotor learning paradigm, although it is unclear that the implicit portion of this paradigm is a result of sensorimotor adaptation.

Explicit and Implicit Learning Processes in DVF

Based on the results of experiment 1, it is clear that this paradigm does not drive only implicit forms of learning, as suggested by previous studies. This is supported by the size of the aftereffects, which are typically thought to be proportionate to the amount learned in sensorimotor adaptation (Shadmehr and Mussa-Ivaldi 1994). During the Catch Trial, the aftereffects were much smaller than expected. In experiment 2, we examined if the DVF locomotor learning paradigm has an explicit component by changing the verbal cues provided. Based on these results, we conclude that the DVF learning paradigm is dominated by explicit strategic learning but contains a portion of implicit learning that remains despite verbal cues to return to baseline walking. There are two pieces of evidence from the present study that support this assertion.

First, with instructions, subjects in the INS Post group in experiment 2 were able to recall and repeat the pattern learned, demonstrating a clear switch between patterns with a verbal cue (Figs. 3B and 5B). The provision of a verbal cue to continue using the developed strategy was needed to prompt the use of that strategy, as evidenced by the results of experiment 1 in which subjects did not receive the verbal cue and thus returned toward baseline symmetry during the Catch Trial. This impact of verbal instruction on the use of a newly learned behavioral strategy has also been found in other tasks (Hussain et al. 2013; Morehead et al. 2015). Hussain et al. (2013) found that, when healthy individuals learned a walking pattern with a new swing duration through DVF, they were able to continue using that pattern when instructed to do so. This supports our results and suggests that verbal cues can result in changes in the observed behavior following locomotor learning with DVF. The results of Morehead et al. (2015) also suggest that verbal cues impact the use of strategy. In that study, they assessed savings, or the ability to relearn a task more quickly during a second exposure to the same task, using an upper extremity visuomotor rotation task. Seven reaches into the second learning phase, subjects were told that the rotation would no longer be present and were instructed to reach directly for the target. They found that, when explicitly prompted to aim to the target, subjects stopped using the learned strategy and reached to the target with a small amount of error (Morehead et al. 2015).

The second piece of evidence suggesting that the learning that occurred during the DVF locomotor paradigm is a combination of both explicit and implicit learning processes is the significant difference between the Catch Trial in the INS group from Baseline (experiment 1) and the trend toward a difference between Baseline and Normal Walking in INS Post (experiment 2). This result suggests that there is a small portion of implicit learning that occurs with this task that is observed during the Catch Trial in experiment 1 and during Normal Walking in experiment 2 (Fig. 5C). In Morehead et al. (2015), the authors suggest that the small amount of error that remained after verbal instruction to aim directly for the target represented the amount of implicit sensorimotor adaptation that occurred during the visuomotor rotation task while the change in hand angle that occurred represented the strategy that was developed. This has also been suggested in a study by Taylor et al. (2014) in which subjects were asked to report the direction of their aim during a visuomotor rotation task. They found that the reported direction of aim did not exactly match their actual hand position, suggesting that a portion of the behavior change was the result of strategy, represented by the reported aiming, and another portion was because of implicit sensorimotor adaptation, represented by the difference between the reported aim and the actual hand position (Taylor et al. 2014). Similar to these results, when our subjects were not prompted to use the developed strategy (e.g., were instructed to just walk normally), we observed a sharp drop in step asymmetry but not a complete return to baseline. We propose that the remaining asymmetry represents the implicit component of the learning in this paradigm; however, unlike the visuomotor rotation task, the implicit portion that remains could be the result of sensorimotor adaptation or use-dependent learning.

Implicit Learning during DVF

Interestingly, the implicit component of learning was not different between any of the groups. It is possible that the similarly sized aftereffects occur because there is a maximum amount of implicit sensorimotor adaptation that can occur with the DVF locomotor paradigm and that all of our groups reached that point. This possibility is supported by past research using error clamping with upper extremity visuomotor rotation tasks. Specifically, Morehead et al. (2017) found that groups who were allowed to develop a strategy during the visuomotor reaching task and those who were not allowed to do so because of error clamping demonstrated similar aftereffects (Morehead et al. 2017). The authors suggest that sensorimotor adaptation has a “ceiling” that was reached at this point. Thus, this is one possible explanation for the lack of difference between the amount of implicit learning that occurred in our groups.

However, there are characteristics of the observed aftereffects that do not align with what we know about sensorimotor adaptation. For example, sensorimotor adaptation decays over time (i.e., deadapts); however, there is no decay noted during Normal Walking in experiment 2 or the Catch Trial in experiment 1. Therefore, it is possible that the implicit portion of this paradigm is not sensorimotor adaptation but rather use-dependent learning. Use-dependent learning results in a direction bias of future movements toward a previously practiced direction even when the perturbation is no longer present (Haith and Krakauer 2013; Huang et al. 2011; Verstynen and Sabes 2011). This form of learning does not require a sensory prediction error like sensorimotor adaptation; however, use-dependent learning can occur with other learning processes (Diedrichsen et al. 2010; Verstynen and Sabes 2011). Based on the experiments conducted here, we are unable to conclude if the aftereffects observed during the DVF locomotor learning paradigm are because of use-dependent mechanisms or sensorimotor adaptation. Thus, although the explicit portion of this learning paradigm is consistent with strategic learning, the implicit portion could reflect contributions from sensorimotor adaptation or use-dependent plasticity.

Limitations and Future Directions

Despite our findings, this study was not without limitations. One limitation is that the distortion in this study was provided only gradually. Given that the perturbation schedule (gradual vs. abrupt) impacts learning, it would be interesting to see how individuals respond to an abrupt perturbation in this DVF paradigm (Roemmich and Bastian 2015; Torres-Oviedo and Bastian 2012). Another limitation is that the sample size for the INS Post group in experiment 2 is small (n = 6); however, given the magnitude of the effect size (see experiment 2 in results), we believe that our sample size was sufficient. Additionally, we are unable to conclude what forms of implicit learning occur in the DVF locomotor learning paradigm with the experiments in this study. Additional studies that include walking bouts with veridical feedback and varied durations of practice would allow for further examination of the implicit learning occurring with this paradigm. Lastly, it is possible that time may impact the contribution of the explicit and implicit component of this paradigm. This possibility is seen during Retention 1 of experiment 2. However, because we did not counterbalance the Retention and Normal Walking phases during experiment 2, the impact of the time cannot be determined from this study. Future work should examine this.

Conclusion

The results of this study suggest that learning during this DVF locomotor paradigm appears to occur primarily through an explicit strategic learning process with a small portion of implicit learning. This is supported by three primary findings. First, in experiment 2, individuals were able to continue to use the learned pattern when prompted to do so even without VF, suggesting that a strategy was developed that individuals could select to use when appropriate. Second, the aftereffects observed during the Catch Trial in experiment 1 were different from baseline walking, suggesting some implicit learning; however, the amount stored was much smaller than the amount learned, which is uncharacteristic of adaptation. Finally, in experiment 2, when asked to return to normal walking, individuals continued to walk with only a slightly asymmetric pattern, suggesting that only a small portion of implicit learning occurred. These final two points suggest that the portion of implicit learning is relatively small compared with the contribution of explicit learning. This is important because previous studies using DVF or split belt locomotion have suggested that locomotor learning relies primarily on implicit learning. Instead, our results suggest that locomotor learning can be dominated by either implicit learning or explicit strategic learning, depending on the specific task used. This DVF locomotor learning paradigm appears to be dominated by explicit strategic learning; thus, it provides a novel learning paradigm to explore explicit locomotor learning. Exploring locomotor learning through a more explicit strategic paradigm is critical given that much of community locomotion requires conscious control of locomotion and that locomotor rehabilitation relies heavily on explicit learning techniques.

GRANTS

This work was funded by the National Institute of Child Health and Human Development Grant 1R01-HD-078330-01A Behavioral and neurophysiologic processes of locomotor learning after stroke and The Foundation for Physical Therapy’s 2016 Florence P. Kendell Doctoral Scholarship.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

M.A.F., S.M.M., and D.S.R. conceived and designed research; M.A.F. performed experiments; M.A.F. and C.C.C. analyzed data; M.A.F., S.M.M., C.C.C., and D.S.R. interpreted results of experiments; M.A.F. prepared figures; M.A.F. and C.C.C. drafted manuscript; M.A.F., S.M.M., C.C.C., and D.S.R. edited and revised manuscript; M.A.F., S.M.M., C.C.C., and D.S.R. approved final version of manuscript.