Role of Haptic Cues in Motor Learning  

Dongwon Kim1 Brandon J. Johnson1 R. Brent Gillespie1 Rachael D. Seidler2 University of Michigan, Ann Arbor, MI, USA

 

ABSTRACT We introduced haptic cues to the serial reaction time (SRT) task alongside the standard visual cues to explore the relative contributions of haptic and visual cues to motor memory and perceptual memory. Motor learning is a complex process that likely depends on the availability of haptic sensory feedback. Motor learning is also often an implicit process, wherein the learner is not even aware that learning is taking place. The SRT task was devised to study implicit learning. Reaction times (RTs) are reduced with practice even in the absence of explicit learning about a sequence embedded in a train of cues. We adopted an SRT protocol developed by Willingham in 1999 to determine whether haptic cues contribute differently than visual cues to the balance of motor and perceptual learning. Experimental results involving 32 participants showed that sequence learning occurs implicitly with haptic stimuli to much the same extent as visual stimuli. Also, it was revealed that the dependence on motor memory (as opposed to perceptual memory) was greater in the group responding to haptic cues than the group responding to visual cues. Comparing the dependence on motor versus perceptual memory, the haptic group reached marginal significance (p=.09), whereas the visual group difference was not significant (p=.23). KEYWORDS: Sequence learning, haptic cue, motor memory, serial reaction time task. Index Terms:

1 INTRODUCTION Motor learning seems to take place without the involvement of conscious or declarative processes. Fitts and Posner called the final stage of motor learning the autonomous phase [1], to indicate that tasks are mastered with practice and not by exercising cognitive or executive functions. To take a real life example, typing a common word on a keyboard is largely an automatic process involving the execution of a learned motor program or sequence. If the keys on a keyboard were re-arranged, conscious involvement might be required in the initial phases of re-learning to type, but eventually new automatically generated sequences would emerge given motor practice. Interestingly, beyond being uninvolved in motor learning, the conscious mind is often not even aware that learning is taking place. The term implicit learning is used to describe this lack of awareness [2]. Now, a second observation regarding motor learning is in order here: the process of acquiring motor skills is often accompanied

by mechanical interaction with the environment. Much of the stimulus that signals success or alerts for necessary adjustment during motor learning comes in the form of haptic stimulus. The haptic stimulus may even become a necessary feature or part of the motor program. Its absence might break the program. Several studies have investigated the efficacy of manual

guidance delivered by a haptic device for training motor skills [3-6]. Haptics might even hold advantages over other modalities given the rather intimate role that haptic stimulus plays in the development and execution of motor programs, as noted above. The appropriate haptic stimulus, if it could be synthesized during interaction with a virtual environment, might activate motor memory without the participation of declarative systems. Many of these studies have, however, reported inconclusive results and generally indicated that motor learning depends in very complicated ways on haptic stimulus. It is probably safe to say that little basic understanding of motor learning has resulted to date from studies involving a haptic device. We do not yet understand how haptic information expedites learning or how it affects the nature of the acquired skill representation. While not involving haptics, nor necessarily focusing strictly on motor skill, several very powerful experimental paradigms about learning and memory have been developed in the psychology literature. One such paradigm is the Serial Reaction Time (SRT) task [7]. In a the standard SRT task, a sequence of stimuli (usually presented visually) consisting of 10 or 12 items that is difficult to memorize explicitly is presented repeatedly while participants press spatially corresponding keys in response. Reaction times (RTs) are typically faster for sequenced than unsequenced blocks of trials, indicating that participants gain knowledge of the sequence during task performance. Of note is that in SRT tasks, participants are often unable to express their knowledge of the sequence, indicating that learning occurred implicitly. Diverse variants of the SRT task have been developed in an attempt to distinguish the contributions of motor memory versus perceptual memory to the learning process (See [8] for a review). In the current study, we introduced haptic cues to a SRT task

alongside the standard visual cues and compared sequence learning outcomes. Most previous SRT studies involved only visual stimuli [e.g. 9, 12], though a few studies employed audio or vibrotactile stimuli [10, 11]. Our haptic stimulus, rendered through motorized keys, was designed to induce significant motion in the fingers. We presumed that by back-driving the tendons and joints of the fingers, we might induce or reinforce motor memory, given the excitation of similar kinesthetic cues during the response key presses; even in visual SRT tasks, there is evidence that sequence knowledge is acquired primarily through motor memory reinforced through key-pressing motions [e.g. 12]. To gain insight into whether or not haptic stimuli contribute

more to motor-based or perceptually-based learning, we applied the experimental protocol developed in [12]. This protocol tests whether participants learn the sequence based on the response

1Department of Mechanical Engineering. {gloryn, bjjohns, brentg}@umich.edu. 2Department of psychology, School of Kinesiology.

rseidler@umich.edu.

265

IEEE World Haptics Conference 201314-18 April, Daejeon, Korea978-1-4799-0088-6/13/$31.00 ©2013 IEEE

 

                 

Figure 1. A schematic diagram of the experimental apparatus and setup.

 

(a)

(c)

(b)

sequence (motor-based) or based on the stimulus sequence (perceptually-based). We expanded the protocol to include both haptic and visual cues. We hypothesized that the haptic modality is more closely tied to the motor system and would therefore support the expression of greater motor-based versus perceptually-based learning.

2 METHODS Apparatus Four lever-shaped keys were motorized to present haptic stimuli using flat voice coil motors (VCMs) and instrumented to record participants’ responses. Every VCM consisted of two back-irons each affixed with four magnets. An aluminum hub was wrapped with 28 AWG magnet wire and this subassembly was set in bearings on a steel dowel between the two back-iron/magnet assemblies. Each key was connected to the top of an aluminum hub, and this rotary assembly had a moment of inertia of 4500 kg·mm2 about its axis of rotation. The displacement of each key was measured by an optical encoder with a resolution of 0.18 degrees. The encoder module was fixed to the bottom of a back-iron for reading a code strip attached along the bottom edge of the aluminum hub (see Figure 1). Each of the VCMs was connected and powered by an Advanced Motion Controls®

brushless servo drive (California, USA). The servo drives and encoders were interfaced with a Sensoray® 626 Data Acquisition board (Oregon, USA) installed on a standard PC. Each key was 160 mm long and 48.5 mm high from the axis of rotation to its end and its top, respectively (a and b in Figure 1). The travel at the tip of each key when fully depressed from the unloaded state to a physical keybed was 12 mm (c in Figure 1). The four keys were equally spaced, and the distance between key centers was 65 mm. To indicate to participants when their responses had been

registered by the computer, the mechanical behavior of each key included a detent or click-feel. This type of keyswitch provides feedback when a response is captured, and is preferred because it supports fast typing speeds and low error rates [13]. Figure 2 shows the force-displacement curve programmed for each key. Let 𝑥 denote the displacement of a key from its equilibrium position. The fixed displacement xA=5mm was defined as a threshold past which participants would have to press the key to register a response to the computer. The displacement xA was also used to measure the reaction time (see Figure 3) and defined the “make-force” (the maximum force past which the key would “break through” to keybed. The force was programmed using simple proportional feedback control with a

266

 

gain 𝐾1=0.22   N/𝑚𝑚. The commanded force increased in proportion to the displacement x from equilibrium such that the make-force was 1N. Once the key crossed the threshold xA, a distinct drop in force occurred during key travel, and a constant force (0.1N) was applied so that the key would return to the equilibrium position if released. Beyond xB , an additional force with a higher effective stiffness (𝐾2=0.6N/𝑚𝑚) acted. The physical keybed was located 12 mm below the equilibrium position. Care was taken to ensure that all four keys provided the same sensation when depressed. Stimuli to elicit participants’ responses in the SRT task were

presented in two ways: visual and haptic. Visual stimuli were provided by the lighting of four horizontally positioned red LEDs spaced 5 cm apart against a black background situated about 90 cm from the participant’s eyes (visual angle: 9.5 degrees). Haptic stimuli were generated by injecting an upward 100 ms half-sinusoid pulse in the reference position of the key, with an amplitude of 7 mm with respect to the tip of the key. Since proportional control was also used with the effective stiffness 0.22 N/mm for haptic stimuli, a corresponding pulse in force and an associated pulse excursion in position would be delivered to a finger. The haptic stimuli were delivered to the fingers of the ring and

index fingers of both hands resting on the keys. Each finger was mapped to a key (labeled 1-4 from left to right) with the left ring finger responding on key 1, left index finger on key 2 and so on. We followed Abrahamse et al. [11] in presenting stimuli to the index and ring fingers instead of the adjacent fingers in order to maximize the ability to discriminate which finger was being stimulated. Reaction time was defined as the difference between the initial

stimulus command and the time at which the threshold  𝑥𝐴 was crossed, as shown in Figure. 3. The participants’ view of their hands was blocked using a box and white noise was presented via headphones in order to eliminate spurious cues.

Participants A total of 32 volunteers (24 male) from the University of Michigan, ranging in age from 19 to 33 years (24.28 ±3.77SD), participated in the study. All participants reported normal or corrected-to-normal vision, and no neurological or motor deficit. None of the participants had previous experience with SRT tasks. All gave written informed consent. The experiment was

approved under the University of Michigan’s Behavioral Sciences Institutional Review Board. Procedure All participants were randomly and evenly divided into two groups: one group (N=16) responded to visual stimuli (the visual group), while the other group (N=16) responded to haptic stimuli (the haptic group). Each of these two groups was randomly and evenly divided into two subgroups. One of the two subgroups (N=8) was assigned to the SRT task under the perceptual condition, whereas the other subgroup (N=8) was assigned to the motor condition. For convenience, the four subgroups will be named visual-perceptual, visual-motor, haptic-perceptual, haptic-motor groups, respectively. The motor and perceptual conditions will be explained below.

Responses to stimuli were made either in the so-called “incompatible” or “compatible” stimulus-response mappings. In the incompatible mapping, participants were instructed to press the key one position to the right of the position at which the stimulus appeared. If the stimulus on the far right appeared, they were to press the key on the far left. In the compatible mapping, on the other hand, participants were asked to press the key that spatially corresponded to the stimulus location. The experiments unfolded in three stages; the familiarization, training, and transfer phases. Before the SRT task began, the familiarization phase was introduced to allow participants to practice making responses on the key apparatus. This phase provided an opportunity to teach participants how to press the keys properly and to ensure they understood the incompatible and compatible stimulus-response mappings. Participants were allowed to view their hands during this exercise to facilitate this learning process. When participants were able to demonstrate proper keypressing and stimulus-response mappings, the familiarization phase was stopped, and the SRT task began. In the training phase, all participants performed the SRT task in

the incompatible stimulus-response mapping, regardless of what kind of stimuli were presented. All the four subgroups experienced the same stimulus sequences during this phase. Shortly afterward, the transfer phase followed to test whether

or not the sequence knowledge acquired during the training phase appeared under the compatible mapping and under specific conditions. In the transfer phase, the visual-perceptual and the haptic-perceptual groups responded using the compatible mapping to cues delivered in sequences that were not altered from those delivered during the training phase (under

Figure 3. The command signals for visual and haptic stimuli and recorded responses. (a) The control signal to turn on/off the LED for visual stimuli. (b) The reference position to be followed by the tip of the key for haptic stimuli. (c) A sample of a typical recorded trajectory responding to a visual stimulus. (d) A sample of typical recorded trajectories in response to a haptic stimulus (dash line: without the finger resting on the key, solid line: with the finger resting on the key).  

Figure 3. The command signals for visual and haptic stimuli and recorded responses. (a) The control signal to turn on/off the LED for visual stimuli. (b) The reference position to be followed by the tip of the key for haptic stimuli. (c) A sample of a typical recorded trajectory responding to a visual stimulus. (d) A sample of typical recorded trajectories in response to a haptic stimulus (dash line: without the finger resting on the key, solid line: with the finger resting on the key).  

(c)

(b)

(d)

(a)

267

 

the perceptual condition). However, the visual-motor and haptic-motor groups responded during transfer using the compatible mapping to cues delivered in sequences that were shifted so that the sequence of responses (key presses) would turn out identical to those made in the training phase (under the motor condition). As it were, while under the perceptual condition, the sequence of responses was “shifted” so that cue delivery (perceived sequences) was the same across the training and transfer phases, under the perceptual condition, the sequence of responses was "unshifted" so that the motor responses were the same across training and transfer phases. Both haptic and visual stimuli were presented for a 100 ms

time interval and then turned off, as described in Figure 3. This differs from most SRT task studies in which stimuli persist continuously until the responses were made [e.g., 11]. However, the authors did not find it necessary to provide stimuli until participants responded. The 100 ms stimulus proved to be enough to draw participants’ responses. A brief pilot study did not indicate the dependence of the RT on duration of stimuli. Response-to-stimulus interval (RSI) was 250 ms for correct responses. All participants were asked to respond as fast as possible without making errors according to the appropriate stimulus-response mapping. Responses were declared erroneous when participants failed to press the appropriate key or make a response within 1.5 s of stimulus presentation. Errors were signaled to participants via audio tones and an extended RSI of 1 s. Thirty second breaks were provided between blocks. Stimulus events (cues) were organized into sequences of 12 and were constructed according to the rules of second-order conditional (SOC) sequences. In a SOC sequence, each event can be predicted only by a unique combination of two preceding events and each pairwise association is equally likely so that pairwise association cannot be used to predict subsequent stimuli [14]. These sequences were then organized into blocks of 108 events. Two types of blocks were presented: sequence blocks that consisted of one SOC (242134123143) repeated nine times, and pseudorandom blocks which consisted of nine distinct, successively presented SOCs picked from a pool of 12. Sequences were presented seamlessly such that participants were only aware of a set of 108 events. The training phase comprised one pseudorandom block

succeeded by 7 sequence blocks, a pseudorandom block (Block 9) and a final sequence block (Block 10). The first pseudorandom block acclimated participants with the task and established a baseline reaction time while the final pseudorandom block allowed us to differentiate sequence learning from general practice effects. Transfer consisted of two pseudorandom blocks (Blocks 11 and 12) for adjusting to the new mapping; one sequence block (Block 13) and a final pseudorandom block (Block 14). A given participant never experienced the same pseudorandom block twice. Median RT and error percentage were displayed for participants between blocks.

Awareness survey After the experiment, a 6-question survey was conducted to determine how much explicit knowledge participants had gained. Question 1 asked participants to choose from a list of four alternatives the statement that best described the task carried out: 1) “Stimulus presentation was completely random”, 2) “Some fingers had to respond more often than others”, 3) “Sometimes I wanted to respond before stimulus presentation”, and 4) “Stimulus presentation was mostly structured” [11].

Questions 2 and 3 were adapted from the Process Dissociation Procedure (PDP) introduced in [15]. Question 2 required participants to generate the 12-event sequence experienced (inclusion) while Question 3 asked participants to generate another 12-event sequence that completely avoided the first (exclusion). Participants were told to recall the sequence as experienced in the training phase. During these exercises, populating the sequence by repeating smaller patterns was not allowed (e.g., 123412341234 would not be a valid response). Thereby, chance level was 0.33. In Question 4, six different SOC sequences were presented through the apparatus in whichever modality participants trained and the participants were asked to identify the correct one (the sequence experienced during training). Questions 5 and 6 asked the participants to rank their engagement in the task and the task’s difficulty on a scale of 1 to 5 (5 being “very engaged” and “very difficult,” respectively). Measures Median RTs were obtained for every sequence repetition per block. The median was then averaged across participants to find the overall RT score per block. The learning measure in the training phase was determined by averaging RTs for Blocks 8 and 10 and comparing that value with the corresponding measure in Block 9. Similarly, in the transfer phase, the learning measure was quantified by finding the difference between the average of Blocks 12 and 14 and the corresponding measure in Block 13. RT data from error trials were excluded from final analysis. To obtain awareness scores, the generated sequences in Questions 2 and 3 were broken into 3-element chunks. The sequence used in that participant’s sequence blocks was likewise divided. Chunks from the generated sequences were compared against those in the actual sequence and the number of correct chunks was divided by 12 (the maximum possible number of correct chunks) resulting in an awareness score between zero and one. Data analysis Data was eliminated if a participant’s error percentage averaged across blocks laid beyond 2 standard deviations of the mean of all participants’ error averaged percentages. Statistical analyses were performed with SPSS (Windows v.18, SPSS Inc.). Mixed analysis of variances (ANOVAs) were performed on median RTs with Stimulus and Condition as between-subject variables and Block as within-subject variables. For analysis of Question 5 and Question 6, two-way ANOVAs were conducted with Stimulus and Condition as between-subject variables and score as a within-subject variable. If the sphericity assumption in ANOVAs was violated, Greenhouse-Geisser adjusted P values were used. t tests were additionally carried out to compare between two values. Significance level was set at 0.05.

3 RESULTS Figure 4 displays the means by subgroup of individual median RTs for the 14 blocks. Downward trends in RT appeared as participants practiced between Blocks 2-8. Since a pseudorandom block followed sequence blocks, increases in RT occurred at Block 9. Notable drops across the training phase to the transfer phase were exhibited, which would be expected because the compatible mapping is easier than the incompatible mapping. Also, as expected, decreases in RT appeared at sequence Block 13 because participants benefited from knowledge of the sequence practiced during the training phase.

268

 

Blocks 2 to 8 An ANOVA result showed a significant main effect of Stimulus [F(1, 28) =12.232, MSE=.515, p < 0.001, 𝜂𝑝2=.304], implying that the haptic group responded slowly in comparison to the visual group. A main effect of Block [F(4.085. 28) =31.434, MSE=.049, p < 0.005, 𝜂𝑝2=.529] was significant. Polynomial contrasts reported a linear trend in Block [F(1, 28) =100.327, MSE=.197, p < 0.001, 𝜂𝑝2=.782], which indicates that the mean RTs decreased over the training phase. The other main and interaction effects did not reach significance (all p>0.1). Learning measure As for sequence learning in the training phase, the ANOVA reported a significant main effect of Stimulus [F(1, 28) =12.908, MSE=.123, p < 0.005, 𝜂𝑝2=.325], reflecting higher RTs for the haptic group compared to those of the visual group. A significant main effect of Block [F(1, 28) =31.89, MSE=.039, p <0.001, 𝜂𝑝2=.532] revealed higher RTs of pseudorandom Block 9 than in surrounding sequence Blocks 8 and 10. The other main and interaction effects were not significant (all p>0.1). One-tailed t tests revealed the differences in RT between the means of Blocks 8 and 10 and Block 9 were significantly greater than zero in all the four subgroups (p<0.05). As for sequence learning in the transfer phase, the ANOVA revealed a significant main effect of Stimulus [F(1, 28) =35.07, MSE=.149, p < 0.001, 𝜂𝑝2=.556], reflecting higher RTs of the haptic group than those of the visual group. A significant main effect of Block [F(1, 28) =75.979, MSE=.012, p < 0.001, 𝜂𝑝2=.731], reported that participants responded fast at sequence Block 13 versus Blocks 12 and 14 indicating that sequence learning was shown across groups. But there was a significant Block×Condition interaction [F(1, 28) =4.431, MSE=.001, p < 0.05, 𝜂𝑝2=.137] implying that the difference in RT between Block 13 and the mean of Blocks 12 and 14 was greater in the motor condition than in the perceptual condition. Additionally, separate independent-sample t tests indicated that there was a marginally significant difference in the leaning measure during the transfer phase between the perceptual and motor conditions

for the haptic group [t(14)=1.808, p=0.09], as depicted in Figure 5. In case for the visual group, there was no significant difference (p=0.23). The other main interaction effects were not significant (all p>0.1). One-tailed t tests revealed the differences in RT between the means of Blocks 12 and 13 and Block 14 were significantly greater than zero in all the subgroups other than the haptic-perceptual group (the haptic-perceptual group: p=0.068).

Awareness Table 1 summarizes the results of the awareness survey. In Question 1, the correct answer is option 4 “stimulus sequence was mostly structured” [13]. As shown in Table 1, the haptic-motor group recorded a low rate for the correct answer in comparison with the other subgroups. In Questions 2 and 3, awareness scores were calculated for

each participant in both inclusion and exclusion recall tasks. An ANOVA revealed no significant main or interaction effects, suggesting that no significant explicit knowledge was obtained in all the four subgroups. In Question 4, all subgroups except for the haptic-perceptual

group had only two members recognize the correct sequence. One participant of the visual-perceptual group was not able to complete this question. The result of the ANOVA on Question 5 revealed no main significant effect and interaction (all p>0.1). In Question 6 about the difficulty of the task, the result of an ANOVA reported a significant main effect of Stimulus [F(1, 28) =9.9, MSE=7.03, p < 0.005, 𝜂𝑝2=.261], indicating that the haptic group showed higher scores than the visual group. Any other main effect and interaction was not significant (all p>0.1).

4 DISCUSSION The objective of this study is twofold: to investigate whether motor learning with the use of haptics could occur implicitly, and whether the haptic modality expedites motor-based learning, perceptual learning, or both. Though several studies have examined motor learning with the

use of the haptics, no study has explored how haptic feedback or haptic cues could be used in motor learning. In particular, the relation between haptic feedback or haptic cues and implicit learning has not been established, even though we would observe that motor learning is mainly implicit and that haptic feedback or haptic cues would seem to be an integral part of motor learning. In the present study, adopting the paradigm of the SRT task, we investigated whether learning with haptic cues is based on declarative memory or based on nondeclarative memory. Furthermore, we explored whether haptic cues engage greater amounts of motor memory relative to perceptual memory and similarly whether visual cues engage motor or perceptual memory to different degrees. The current study demonstrated that haptic cues supported

implicit sequence learning much the same as visual cues, as

visual-perceptual

visual-motor

haptic-perceptual

haptic-motor

Question 1 n=5 n=4 n=5 n=1 Question 2 .458±.099 .417±.126 .469±.117 .417±.204 Question 3 .364±.133 .375±.173 .448±.133 .427±.144 Question 4 n=2 n=2 n=4 n=2 Question 5 4.5±.534 4.625±.518 4.5 ±.534 4.875±.354 Question 6 3.0±.534 2.625±.744 4.0 ±.756 3.5±1.195 Table 1. A summary of awareness survey. The sections of Questions 1 and 4 present how many participants chose the correct answer. The remaining sections present the averaged values by subgroup with standard deviations.

 Figure 4. Mean by subgroup of individual median RTs. Error bars are ± standard error of the mean. R and S stand for pseudorandom and sequenced stimuli, respectively.  

269

 

shown in Figure 4. The awareness survey showed that no significant explicit knowledge was gained across groups. During the training phase, downward trends in RT were exhibited as training advanced from Block 2 to Block 8 in the haptic group as well as in the visual group, indicating that sequence learning was occurring. An even more reliable measure for sequence learning, a difference in RT between pseudorandom Block 9 and the mean of its surrounding sequence blocks, revealed that the extent of sequence learning was similar in the visual and haptic groups. The difference from zero was statistically different in the four subgroups. In general, participants showed slower responses to haptic stimuli than to visual stimuli. The result might originate from the distinct processing systems of visual and haptic modalities that mediate learning, which are involved in the representation and processing of features of perceptual stimuli [8]. Also, the present study demonstrated that haptic stimuli favored motor-based learning over perceptual learning. We examined an indicator of the difference in the RT learning measure between motor-based learning and perceptual learning in the haptic group. The difference of the haptic group reached marginal significance (p=.09), whereas that of the visual group was not significant (p=.23), as presented in Figure 5. It is suggested that the development of a motor representation relative to a perceptual representation is greater in the haptic group than the visual group. In sum, we showed that sequence learning occurs with haptic stimuli, and that haptic stimulus modality enhances motor memory, facilitating motor-based learning in comparison with the visual modality. As haptic stimuli were designed to activate our motor and cognitive haptic system, stimuli as well as responses could be concluded to contribute to creating motor memory for a presented sequence. In the visual SRT task, mainly, responses contribute to creating motor memory for a presented sequence. Several studies have reported on the human’s ability to remember motor patterns that are exploited by haptic training, rather than by visual training [16, 17]. This remarkable memory results from motor memory or kinesthesia, but it could also enable the trainees with haptic cues to consciously remember what they learn, thus increasing declarative knowledge. Our study showed that haptic stimuli in a SRT task facilitated implicit learning, supported by the awareness study suggesting that explicit knowledge was not involved in learning. Furthermore, the present results support the idea that motor learning versus perceptual learning, while

present in both modalities, is stronger with haptic stimuli than with visual stimuli.

ACKNOWLEDGEMENT The authors express our gratitude to Dr. Sile O'Modhrain in the University of Michigan School of Music, Theatre & Dance for her inspiration, insight, and instruction for this study.

REFERENCES [1] P.   Fitts   and   M.   Posner,   Human   Performance.   Belmont,   CA,  

Brooks/Cole,  1967.  [2] A.   Cleeremans.   Mechanisms   of   implicit   learning.   MIT   Press,  

Cambridge,  1993.  [3] T.R.   Armstrong.   Training   for   the   production   of   memorized  

movement   patterns.   Technical   Report   26,   Ann   Arbor,   MI:  University  of  Michigan,  Human  Performance  Center,  1970.  

[4] R.B.  Gillespie,  M.S.  O’Modhrain,  P.  Tang,  D.  Zaretzky,  and  C.  Pham.  The  virtual  teacher.  In  Proceedings  of  the  ASME  Dynamic  Systems  and  Control  Division,  pages  171–8,  1998.  

[5] D.   Feygin,   M.   Keehner,   and   F.   Tendick.   Haptic   guidance:  Experimental   evaluation   of   a   haptic   training   method   for   a  perceptual  motor  skill.   In  Proceedings  of  the  10th  Symposium  on  Haptic   Interfaces   for   Virtual   Environment   and   Teleoperator  Systems  (HAPTICS  ’02),  pages  40–7,  2002.  

[6] F.A.  Mussa-­‐Ivaldi  and  J.L.  Patton.  Robots  can  teach  people  how  to  move  arm.  In  Proceedings  of  the  IEEE  International  Conference  on  Robotics  and  Automation,  pages  300–5,  2000.  

[7] M.J.  Nissen  and  P.  Bullemer.  Attentional  requirements  of  learning:  Evidence   from  performance  measures.   Cognitive  Psychology,   19,  pages  1–32,  1987  

[8] T.  Goschke  and  A.  Bolte.  On   the  modularity  of   implicit   sequence  learning:  Independent  acquisition  of  spatial,  symbolic,  and  manual  sequences.  Cognitive  Psychology,  65,  pages  284–320,  2012.  

[9] D.B.  Willingham  and  K.  Goedert-­‐Eschmann.  The  Relation  Between  Implicit  and  Explicit  Learning:  Evidence  for  Parallel  Development.  Psychological  Science,  10,  pages  531–4,  1999.  

[10] P.  Zhuang,  C.  Toro,  J.  Grafman,  P.  Manganotti,  L.  Leocani,  and  M.  Hallett.   Event-­‐related   desynchronization   (ERD)   in   the   alpha  frequency   during   development   of   implicit   and   explicit   learning.  Electroencephalography  and  clinical  neurophysiology,  102,  pages  374–81,  1997.    

[11] E.L.  Abrahamse,  R.H.J.V.D.  Lubbe,  and  W.B.  Verwey.  Asymmetrical  learning  between  a  tactile  and  visual  serial  RT  task,  The  Quarterly  Journal  of  Experimental  Psychology,  2,  pages  210-­‐7,  2008.  

[12] D.B.  Willingham.   Implicit   motor   sequence   learning   is   not   purely  perceptual.  Memory  &  cognition,  27,  pages    561–72,  1999.  

[13] H.  Brunner  and  R.M.  Richardson.  Effects  of   keyboard  design  and  typing   skill   on   user   keyboard   preferences   and   throughput  performance.  Proceedings  of  the  Human  Factors  and  Ergonomics  Society  Annual  Meeting  October  1984,  28,  pages  3267-­‐271.  1984.  

[14] J.   Reed   and   P.   Johnson.   Assessing   implicit   learning  with   indirect  tests:   Determining   what   is   learned   about   sequence   structure.  Journal   of   Experimental   Psychology:   Learning,   Memory,   &    Cognition,  20,  pages  584-­‐94,  1994.  

[15] A.   Destrebecqz   and   A.   Cleeremans.   Can   sequence   learning   be  implicit?  New  evidence  with   the  process   dissociation  procedure.  Psychonomic  bulletin  &  review,  8,  pages  343–50,  2001.  

[16] F.J.  Clark  and  K.W.  Horch.    Kinesthesia.  In  K.  Boff,  L.  Kaufman,  and  J.   Thomas   (Eds.),   Handbook   of   perception   and   human  performance.  New  York:  Wiley,  1986.  

[17] C.D.   Chapman,   M.D.   Heath,   D.A.   Westwood,   and   E.A.   Roy.  Memory   for   kinesthetically   defined   target   location:   evidence   for  manual  asymmetries.  Brain  and  cognition,  46,  pages  62–6,  2000.  

Figure 5. The learning measure in reaction time. The plots in the training panels represent the values averaged across the perceptual and motor conditions in each group. The third panel shows the values averaged across the visual and haptic groups in each condition during the training phase. Error bars are ±1 standard error of the mean.

 

270