| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
1Brain Sciences Center, Veterans Affairs Medical Center, Minneapolis; and 2Graduate Program in Biomedical Engineering and 3Departments of Neuroscience and Neurology, University of Minnesota, Minneapolis, Minnesota
Submitted 26 July 2006; accepted in final form 27 October 2006
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
; Jordan and Rumelhart 1992; Lackner and Dizio 1994; Shadmehr and Mussa-Ivaldi 1994; Wolpert et al. 1995), which is used to predict the effects of an object or a particular environment on the body and to calculate the motor output required to perform desired actions. The way in which internal models of different environments are acquired has been used to study the characteristics and capacities of motor memory.
Subjects readily learn to operate in new mechanical (Shadmehr and Mussa-Ivaldi 1994) or kinematic (Flanagan and Rao 1995; Krakauer et al. 1999) environments. However, simultaneous exposure to opposite or conflicting force fields (Brashers-Krug et al. 1996; Caithness et al. 2004; Karniel and Mussa-Ivaldi 2002; Shadmehr and Brashers-Krug 1997) or visuomotor transformations (Caithness et al. 2004; Krakauer et al. 1999, 2005; Tong et al. 2002; Wigmore et al. 2002; see Bock et al. 2003, for a different perspective) often interferes with the learning process. Interference has been observed both when the fields were alternated trial by trial or presented in alternating blocks. It is generally believed that the interference observed after exposure to opposite force fields in close temporal proximity is due to disruption in the consolidation of the short-term motor memory (but see Caithness et al. 2004).
This view has been recently challenged by demonstrating that simultaneous learning of conflicting viscous force fields is possible if these fields are presented randomly and appropriate contextual cues are provided (Osu et al. 2004; Wada et al. 2003). To explain this intriguing result, the authors of these studies proposed that the subjects formed two internal models simultaneously and were able to switch between them using the available contextual cues, in accordance with the MOSAIC model (Haruno et al. 2001; Wolpert and Kawato 1998). The MOSAIC model proposes that, during motor adaptation, many controllers (inverse models) can be simultaneously selected and learned. Before performing a task, the a priori information available from the extrinsic contextual cues guides the selection of the appropriate controller and thus facilitates rapid and effective switching.
This finding raised the question whether such simultaneous context-based learning of conflicting environments could be applied to other more general situations as well. One important generalization would be to uncertain environments in which the forces vary randomly in magnitude (Scheidt et al. 2001; Takahashi et al. 2001) as well as direction. Based on these earlier results, we reasoned that given sufficient practice and appropriate contextual cues, subjects should be able to adapt to a combined environment comprising two randomly switching identical distributions of field strengths, one generating clockwise (CW) and the other generating counter-clockwise (CCW) forces. Specifically, they would be expected to compensate for a perturbation close to the mean value of the applied environment (Scheidt et al. 2001; Takahashi et al. 2001). Our results, discussed in the following text, demonstrate that human subjects were not able to form internal models of such an unpredictable and conflicting environment.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
Overview of experiments
The experiments involved subjects performing reaching movements in a single direction against perpendicular velocity-dependent forces that pushed either to the right or to the left, were randomly switched and were cued by two different colors. For the first two experiments, the magnitude of the field strength was stochastic from trial to trial. Experiment 1 was performed over 1 day, whereas subjects in experiment 2 performed the task over two consecutive days to allow them more practice with the fields. Experiment 3 was also performed over 2 days, but in this case, the magnitude of the field strength was fixed to control for the trial-to-trial stochastic variation in the force magnitude.
Experimental setup
For all experiments, subjects sat in a chair holding the handle of a two-joint, planar robotic manipulandum (Interactive Motion Technologies, Cambridge, MA) and executed timed reaching movements in the horizontal plane. A vertical LCD monitor in front of the subjects displayed "start" and "end" target circles, both of 3 cm diameter, and a 1.5 cm-diameter cursor representing the hand (equivalent to the robot endpoint) position; the hand and the arm were visible throughout the experiment. The reach was a 15 cm movement and corresponded to an upward motion of the cursor on the screen and a movement of the subjects' hand away from the body in a para-sagittal plane. Subjects were asked to position themselves so as to approximately align the forward movement with the axis of the shoulder joint. However, their posture and position were not constrained in any way.
On most trials, the robotic device produced forces on the subjects' hand (see following text for details). The forces were applied only during the outward movement from the start to the end target, and subjects were allowed to return to the start circle however they chose (the start circle was visible at all times). A trial began when the subject positioned the cursor completely within the start circle. On force trials in all three experiments, a colored rectangular frame was displayed around the workspace on the monitor at the beginning of the trial, as a contextual cue. For experiment 1, the frame was blue for rightward forces and red for leftward forces (vice versa for the other 2 experiments). For the initial "null trials" in the absence of forces, a cue was irrelevant and was therefore not provided. After the cue was displayed for 300 ms, the end target came on following a random delay of 200–500 ms. Subjects were instructed to wait for the target to come on and then make point-to-point reaching movements that were "as straight and smooth as possible" and to "stop and hold at the target." During force trials, the forces were applied once the subject crossed a circular window that was 1.5 times the radius of the start circle and concentric with it (except in experiment 3; see following text). The movement was required to be completed within 400–700 ms; this relatively large time window was used because of the wide variation in the magnitudes of the field strengths that were used. A trial was complete after a target hold time (700 ms). At the end of each trial, feedback was provided regarding the movement timing: the target turned red if the movement was too slow, turned green if it was too fast, and turned black if subjects did not stop and hold at the target for the required time. After that, the trajectory of the movement was shown as feedback to the subjects for 300 ms.
Incorrect movements were classified into four types: limit errors, when the subjects' hand strayed outside a virtual boundary in the workspace (the extent of the virtual boundary was the same as the movement amplitude in the direction of the movement and 1.5 times the movement amplitude in the direction perpendicular to the movement); fast and slow errors, when the movement was <400 ms and >700 ms respectively; and target hold errors, when the subject did not stop and hold at the target for the required time.
Experiment 1
Six subjects (1 male, 5 females) participated in experiment 1. The experiment consisted of 27 blocks of 20 trials each performed in a single session. A break of 12 s was provided after each block.
Subjects started with a block of reaching movements under full on-line visual feedback, in the absence of forces (the null field), to get them acquainted with the apparatus. This was followed by another block without the forces but now without on-line visual feedback of the cursor on the display monitor. This was accomplished by blanking out the cursor when a window, of radius 1.5 times the radius of the start circle, was crossed. After these two blocks in the null field, subjects were asked to perform 25 blocks in the presence of forces without real-time visual feedback of the cursor (similar to the 2nd block). The real-time feedback of the cursor was suppressed to avoid on-line error corrections during the movement.
The forces were generated by two torque motors acting on the handle of the robotic arm, which the subjects held with their right hand. The forces were proportional to the hand velocity and directed perpendicular to the desired direction of motion; i.e., the forces tended to perturb the subject's hand either to the left (left-directed force field, LF) or to the right (right-directed force field, RF). The force fields can be expressed as
 | (1) |
represents the y component of the hand velocity, Fx and Fy are the x and y components of the force, and b is the gain coefficient. This gain coefficient was picked at the beginning of each trial, randomly (i.e., with a probability of 0.5) from either of two Gaussian distributions of identical variance (22 N2/m2/s2) and respective means +15 and –15 N/m/s. The gain coefficients were selected right before each trial without any sort of constraint (on number or order). As a consequence, the sequence of coefficients was different for individual subjects (in contrast, for example, with Scheidt et al. 2001). Experiment 2
In experiment 2, six subjects (5 males, 1 female) performed exactly the same task as in experiment 1 with the addition of a second session of 25 force blocks on the following day (a separation of 
24 h). This was followed by two more blocks in which the forces were not applied on certain trials (catch trials), randomly chosen with a probability of 0.25. The catch trials were never allowed to occur consecutively, irrespective of whether a trial was completed correctly or not, to ensure that they maintained the quality of surprise. Furthermore, for the catch trials, the cues were presented as for the force trials so that the subjects could not predict their occurrence. One subject (male) showed hand-path deviations in only one direction (for both the left- and right-directed forces) and was excluded from further analyses.
Experiment 3
Ten subjects (7 females, 3 males) participated in a third experiment, which had the same structure as the second one with two main differences. First, the magnitudes of the gain coefficients were kept fixed at one of three different levels (16, 20 and 23 N/m/s) for individual subjects instead of selecting from Gaussian distributions (the direction of the force was still random), and second, the forces were turned on when the subject entered the center circle, instead of waiting till the subject exited the center as in the previous experiments. Because the forces are proportional to the velocity, subjects would not have felt much force until they started their movement. This is in accordance with the usual paradigm to study movements in viscous force fields. Three subjects (2 females, 1 male) were excluded from further analyses: one was not perturbed, one was perturbed in only one direction and the third showed inconsistent behavior.
Data analysis
To assess the performance of the subjects, we calculated an average signed hand-path deviation for each trial by first computing a sample-by-sample difference between the hand trajectory of the subject and the straight line connecting the start to the end target and then averaging it over the movement (Eq. 2). This approach was motivated by several studies that have shown the straight-line path to be the "desired" trajectory of such reaching movements and that the hand-paths of subjects deviate in the direction of the force (Gandolfo et al. 1996; Morasso 1981; Shadmehr and Mussa-Ivaldi 1994).
 | (2) |
To analyze the trend of the deviations over the course of the experiment, we divided the trials in the force blocks into bins of 50. In each such bin, trials performed in RF were separated from those in LF and then pooled over all the subjects in a particular experiment. These were then averaged to calculate the average signed hand-path deviation for the group. For the null trials, one average was calculated for each of the two blocks (of 20 trials each). To analyze the catch trials, we first assigned them to LF or RF according to the color of the cue that was displayed before the trial and then averaged the deviations on these trials over the subjects. We also calculated the average by pooling all the catch trials together, ignoring the color of the cue.
To study the relationship between the hand-path deviation and the applied field strength for the first two experiments, we binned the gain coefficient values in 2 N/m/s bins and calculated the mean deviation in each bin. These were then plotted against the gain coefficients at the bin centers (Fig. 3).
|
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
We initially asked a group of six subjects to perform a single session of reaching movements during which their arm was perturbed by two randomly presented conflicting viscous force fields (experiment 1). The gain coefficient was selected randomly at the beginning of each trial from either of two Gaussian distributions (Fig. 1A; see METHODS for details). Figure 1B depicts the hand paths of a typical subject from the group during an early (left) and a late force block (right). As expected, the initial trajectories of the subjects deviated in the direction of the forces in both fields. However, the pattern of deviation persisted throughout the experiment so that the hand paths were still curved at the end of the experiment; in fact, for some subjects, the curvature of the trajectory appeared to increase over time. This indicates that the subjects were unable to adapt to the applied forces.
|
|
The points on the far right in Fig. 2B show the average deviation on the catch trials. The points shown in red and blue were calculated by separating trials on the basis of the color of the cue provided (which would indicate the expected direction of the force) and then taking averages. The point in black was calculated by pooling all the catch trials together irrespective of the cue. There is clearly no aftereffect for RF (red), and the one for LF (blue) is not significantly different from zero (P = 0.1142, 2-tailed t-test).
We found similar results using as error measure the deviation at peak y-velocity, normalized by the velocity. Further, to consider only the feed forward component of the movement, we calculated the signed hand-path deviation using only the first 250 ms of the movement and found similar results.
Variation of hand-path deviations with the applied forces
Having not found any signs of adaptation to the force fields, we wondered if there was any pattern in the variation of the deviations with force magnitude, which was proportional to the gain coefficient. As expected, the hand-path deviation varied almost linearly with the gain, both in LF and RF, with higher forces producing proportionally higher deviations (Fig. 3). A simple linear regression of the deviations on the gain coefficients (LF and RF combined) yielded a R2 between 0.65 and 0.83 for experiment 2 (mean ± SD: 0.75 ± 0.07, n = 5). For experiment 1, the corresponding range was [0.38, 0.87] (0.65 ± 0.19, n = 6). To estimate the value of the field strength that was "best compensated for," we calculated the gain coefficient at which the deviation was equal to the baseline control (calculated from the 2 null blocks at the start of the experiment). This was found to be inconsistent among the subjects and, in particular, was not always close to the mean. Specifically, we first compared the baseline-crossing gain coefficient of each individual subject with the mean value of the gain actually experienced by the subject in LF and RF separately. Because the linear regression for LF and RF separately did not yield sufficiently high R2 values (all less than 0.5), we calculated the baseline-crossing gain coefficient by visually inspecting the interpolated curves (as shown Fig. 3). Except for one subject (S2, experiment 1; P = 0.0486) in the LF, the baseline-crossing gain was always significantly different from the mean of the actual set of gains in LF as well as RF (P < 0.001, 2-tailed t-test). Only two other subjects could be said to be close to the mean (S4, experiment 1, LF; S6, experiment 2, LF). If the subjects really were to "learn" the forces for a particular gain coefficient, one would expect the deviation at that gain to be close to the baseline. Thus the preceding results suggest that the subjects did not learn the mean value of either of the distributions, which is in contrast to what might have been expected if the results of some previous studies could be generalized to our set-up (see DISCUSSION). The lack of consistency among the subjects in the baseline-crossing value does not permit us to attach any significant meaning to it. For the same reason, we did not do any comparison of the averages across subjects.
Next, we combined the data from LF and RF together to see if the subjects, not being able to separate the left and the right forces, instead learned the mean of the combined distribution. In this case, we calculated the baseline-crossing gain coefficients from the regression lines. Again, no consistent pattern was found and, except for one subject (S3, experiment 2; P = 0.8), the baseline-crossing gain coefficient was always significantly different from the actual mean gain experienced by the subjects (P < 0.001). Two other subjects in experiment 2 and one in experiment 1 came close to the means. The averages over the subjects were 3.64 ± 4.08 N/m/s (n = 5) for experiment 2 and –10.7 ± 8.84 N/m/s (n = 6) for experiment 1.
Finally, we repeated the preceding analyses using the gain coefficient values at which the deviation was equal to zero, rather than the baseline for each subject, and found similar results.
Another independent measure of performance
One might argue that the force distribution may have been a little too complex, making the task too difficult for the subjects. This may, then, explain the lack of improvement we noted in the performance of the subjects. To test this, for the second experiment we used an additional independent measure of performance—the number of incorrect movements that a subject makes during the course of the task. Figure 4 plots the number of four different types of incorrect movements (see METHODS) averaged over the six subjects. It can be seen that the numbers progressively decline during the course of a single day as well as for the overall experiment. A peak is observed at the start of the second day, but the magnitude of this peak is substantially lower than that at the start of day 1. Indeed, for all the subjects pooled together, the proportion of incorrect movements in the first two blocks of day 1 was significantly greater than that in the first two blocks on day 2 (difference of proportion test, z = 7.0824, P < 0.0001), indicating that there was a consolidation of whatever the subjects learned on day 1.
|
Role of variation in force magnitude
Even though we have shown that subjects were unable to adapt to the conflicting force fields that had both unpredictable direction and magnitude, it was not completely clear what may have been the factor that prevented learning. Because an important novel aspect of our task was the random variation in force magnitude, to isolate its effect, we repeated experiment 2 with the modification that the gain coefficients were of fixed magnitude instead of being selected from a Gaussian distribution. Figure 5 shows the time course of the signed hand-path deviations averaged over seven subjects. Modest "adaptation" seems to have occurred in that the hand-path deviation decreased with time. However, in general, aftereffects were found to be absent. This suggests that instead of acquiring internal models of the conflicting fields, the subjects may have adopted some other strategy (e.g., stiffening the arm) that enabled them to effectively compensate for the applied perturbation. This appears plausible because adaptation with only one movement direction need not require global adaptation and therefore can be accomplished without learning the force field in its generality.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
It has been shown repeatedly in previous studies that exposure to conflicting dynamic force fields in close proximity results in interference, such that no net performance improvement is observed (Brashers-Krug et al. 1996; Caithness et al. 2004; Karniel and Mussa-Ivaldi 2002; Shadmehr and Brashers-Krug 1997). However, subjects do appear to simultaneously learn multiple dynamic fields of similar construction within different state spaces (Fukushi and Ashe 2003). The exact mechanism of the interference seen with conflicting dynamic fields is not clear, but it is generally believed that exposure to a second conflicting field results in retrograde interference leading to disruption of the short-term motor memory associated with the first (Brashers-Krug et al. 1996; Shadmehr and Brashers-Krug 1997) unless sufficient time has elapsed for consolidation of the first field to occur (for another view, see Caithness et al. 2004). Recent work (Osu et al. 2004; Wada et al. 2003), however, has challenged the belief that simultaneous learning of internal models of conflicting force fields is not possible by showing that if the force fields are presented in a random order (vs. a predicable sequence) using appropriate contextual cues in each trial, then learning does appear to occur. The authors explained this as a context-based switching of multiple internal models, although it was not clear what exactly the role of random presentation was and why such context-based switching was not possible under predictable exposure conditions (in alternating trials, for example). Our experiment was designed to test whether their observation could be generalized to more complex dynamic environments in which the conflicting fields could have a variety of different magnitudes, as one might experience in everyday life. Our experimental set-up was, in principle, very similar to that of Osu and colleagues (2004) except for the important difference that we sampled from a distribution of gain coefficients for each field (for the first two experiments), whereas in the task of Osu and colleagues, two fixed coefficients were used to generate the CW and CCW force fields. It has been shown (Scheidt et al. 2001; Takahashi et al. 2001) that subjects tend to learn approximately the mean value of force distributions, if applied only in one direction, both for uni- and bimodal distributions. We wondered if such learning would be observed even when the force distributions are presented simultaneously, randomly in opposite directions, as would be expected if the results of Osu and colleagues were generalizable.
The subjects in our first two experiments failed to show any learning of the conflicting force fields. We also investigated whether the subjects learned the distribution of forces even though they did not acquire an internal model of the fields themselves. Learning of the magnitude could be accomplished in one of two different ways. Subjects might learn the approximate mean of both the "left-centered" and the "right-centered" distribution of gain coefficients. However, if subjects were not able to separate the conflicting force fields, they might treat the distribution of magnitudes to the right and the left as being part of one field in which case they would tend to learn the approximate mean of the combined (bimodal) distribution (theoretically zero). In any case, the hand-path deviation of the subjects in trials in which the learned force strength was applied would be expected to be close to the baseline (control) deviation (or zero). The baseline-deviation gain values (as well the zero-deviation values) that we calculated were in general not found to be close to the appropriate means and thus provide evidence against either form of magnitude learning on the part of the subjects.
The efficacy of contextual cues in enabling subjects to simultaneously adapt to, and predictively switch between, multiple conflicting environments is a matter of debate. Some studies have shown that the simultaneous learning of opposite environments, with appropriate switching (presumably of the two internal models), is possible if appropriate postural (Gandolfo et al. 1996) or spatial (Rao and Shadmehr 2001) cues are provided, but not with arbitrary color cues (Gandolfo et al. 1996), even after extensive training (Shadmehr et al. 2005). Krouchev and Kalaska (2003), however, demonstrated that switching between opposite viscous fields based only on color cues is possible if monkeys are recalling a previously learned task (after extensive practice). One common aspect of the training schedule in these studies is that they used predictable sequences of fields, either trial-by-trial switching or alternating blocked presentation (except Shadmehr et al. 2005, where the authors presented the cue-field associations in a random order). In contrast, other studies (Osu et al. 2004; Wada et al. 2003) demonstrated that switching based entirely on color cues is possible only if the fields are presented randomly. Our results show that, for two opposite distributions of field strengths, switching based only on color cues is not possible, even with random presentation.
To isolate the significance of the unpredictable variation in the field strength, we repeated our experiment using fixed gain coefficients. With only this modification, subjects showed a decrease in the hand-path deviation over time (Fig. 5), suggesting that the primary feature that prevented the subjects from adapting to the forces and improving their performance in the first two experiments was the randomly varying field strength. However, the aftereffects in this experiment were not significant, indicating that, even under the condition of fixed field strength, subjects were unable to form internal models of the conflicting dynamic environments. It is possible that the subjects used impedance control (Franklin et al. 2003; Takahashi et al. 2001) as a strategy for behavioral improvement in experiment 3, given the fixed field strengths, as opposed to the other two experiments in which the magnitude of the fields varied.
Some potential reasons for the discrepancy between our results and those of Osu et al. (2004) and Wada et al. (2003) could be the differences in the saliency of the cues and the number of training days. The color cues we used were displayed only for a short time (300 ms) as opposed to 1.5- 3.5 s in Wada et al. (2003) and 2 s in Osu et al. (2004). We chose a short interval to decrease the total amount of experimental time for the subjects and thus avoid boredom. However, we believe that the short cue interval cannot account for our findings. This is supported by the results of experiment 3 in which subjects did show adaptation with exactly the same cue structure. As regards the period of training, although we had more than 1,000 trials in one direction, the training was done over only 2 days, and it might be argued that subjects may be unable to consolidate complex fields within that time frame. However, Scheidt et al. (2001), using a very similar experimental setup to ours except that the perturbation was applied only in one direction (no conflict), demonstrated the formation of internal models despite the fact that they used a lot fewer trials (400 trials for the bimodal distribution, 200 for the unimodal distribution). Furthermore, subjects in experiment 2, who had an overnight break that might have been expected to consolidate their learning, did not fare better than subjects who practiced for only 1 day (see Fig. 2). Therefore, although we cannot know for certain whether subjects would develop internal models of the fields we used if they practiced for several hours daily over a period of months, we feel confident in claiming that subjects do not learn the fields using training criteria that have been used in many other experiments. Perhaps the most plausible explanation for the discrepancy between our findings and those of Osu et al. (2004) is that the development of internal models in their experiment may have been direction dependent. If one inspects the single-subject data (Fig. 2) in their experiment, no aftereffects are evident in the single direction we used in the current experiment. It is possible that the development of internal models of conflicting fields using random presentations and contextual cues may be influenced by limb biomechanics.
Although subjects did not develop internal models, the amount of exposure they received to the task did enable them to learn some aspects of the behavior. This is seen in the plots in Fig. 4, which clearly show an improvement in performance in terms of a reduction in the number of incorrect movements as the task progressed. This was observed on both experimental days. We also observed an increase in the number of errors at the beginning of the second day, and, interestingly, this number was significantly smaller than that at the start of the first day. We interpret the reduction in the number of incorrect movements as indicating that subjects really did learn some aspects of the behavior although in a different "dimension" from the one of interest. The formation of an internal model likely involves a number of interacting processes working simultaneously, and it is possible that these different processes follow different learning dynamics.
In conclusion, human subjects were unable to form internal models of conflicting dynamic environments that were presented randomly in the presence of contextual cues, even after practice on more than a thousand trials. In a generalized setting, where both the force magnitude and the direction were unpredictable, subjects showed no signs of being able to adapt their hand paths, whereas limited adaptation was seen in exactly the same setup when the field strengths were kept fixed. However, our data show no evidence supporting the formation of internal models of the conflicting fields in either condition. Although it is possible that the amount of practice may not be sufficient to allow learning, this by itself is not sufficient to account for the discrepancies between our findings and those of Osu et al. (2004) and Wada et al. (2003). Rather, our results indicate that the effects of random presentation and context-based switching documented in these latter studies may not apply to more general settings.
|
GRANTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
Address for reprint requests and other correspondence: J. Ashe, Brain Sciences Center (11B), VAMC, One Veterans Dr., Minneapolis, MN 55417 (E-mail: ashe{at}umn.edu)
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONGRANTSREFERENCES |
|---|
Brashers-Krug T, Shadmehr R, Bizzi E. Consolidation in human motor memory. Nature 382: 252–255, 1996.[CrossRef][Medline]
Caithness G, Osu R, Bays P, Chase H, Klassen J, Kawato M, Wolpert DM, Flanagan JR. Failure to consolidate the consolidation theory of learning for sensorimotor adaptation tasks. J Neurosci 24: 8662–8671, 2004.
Flanagan JR, Rao AK. Trajectory adaptation to a nonlinear visuomotor transformation: evidence of motion planning in visually perceived space. J Neurophysiol 74: 2174–2178, 1995.
Flanagan JR, Wing AM. The role of internal models in motion planning and control: evidence from grip force adjustments during movements of hand-held loads. J Neurosci 17: 1519–1528, 1997.
Franklin DW, Osu R, Burdet E, Kawato M, Milner TE. Adaptation to stable and unstable dynamics achieved by combined impedance control and inverse dynamics model. J Neurophysiol 90: 3270–3282, 2003.
Fukushi T, Ashe J. Adaptation of arm trajectory during continuous drawing movements in different dynamic environments. Exp Brain Res 148: 95–104, 2003.[CrossRef][ISI][Medline]
Gandolfo F, Mussa-Ivaldi FA, Bizzi E. Motor learning by field approximation. Proc Natl Acad Sci USA 93: 3843–3846, 1996.
Haruno M, Wolpert DM, Kawato M. Mosaic model for sensorimotor learning and control. Neural Comput 13: 2201–2220, 2001.
Jordan MI, Rumelhart DE. Forward models: supervised learning with a distal teacher. Cogn Sci 16: 307–354, 1992.
Karniel A, Mussa-Ivaldi FA. Does the motor control system use multiple models and context switching to cope with a variable environment? Exp Brain Res 143: 520–524, 2002.[CrossRef][ISI][Medline]
Krakauer JW, Ghilardi MF, Ghez C. Independent learning of internal models for kinematic and dynamic control of reaching. Nat Neurosci 2: 1026–1031, 1999.[CrossRef][ISI][Medline]
Krakauer JW, Ghez C, Ghilardi MF. Adaptation to visuomotor transformations: consolidation, interference, and forgetting. J Neurosci 25: 473–478, 2005.
Krouchev NI, Kalaska JF. Context-dependent anticipation of different task dynamics: rapid recall of appropriate motor skills using visual cues. J Neurophysiol 89: 1165–1175, 2003.
Lackner JR, Dizio P. Rapid adaptation to Coriolis force perturbations of arm trajectory. J Neurophysiol 72: 299–313, 1994.
Morasso P. Spatial control of arm movements. Exp Brain Res 42: 223–227, 1981.[ISI][Medline]
Osu R, Hirai S, Yoshioka T, Kawato M. Random presentation enables subjects to adapt to two opposing forces on the hand. Nat Neurosci 7: 111–112, 2004.[CrossRef][ISI][Medline]
Rao AK and Shadmehr R. Contextual cues facilitate learning of multiple models of arm dynamics. Soc Neurosci Abstr 302.4, 2001.
Scheidt RA, Dingwell JB, Mussa-Ivaldi FA. Learning to move amid uncertainty. J Neurophysiol 86: 971–985, 2001.
Shadmehr R, Brashers-Krug T. Functional stages in the formation of human long-term motor memory. J Neurosci 17: 409–419, 1997.
Shadmehr R and Mussa-Ivaldi FA. Adaptive representation of dynamics during learning of a motor task J Neurosci 14: 3208–3224, 1994.[Abstract]
Shadmehr R, Donchin O, Hwang E-J, Hemminger SE, Rao A. Learning dynamics of reaching. In: Motor Cortex and Voluntary Movements, edited by Vaadia E. Boca Raton, FL: CRC, 2005, p. 297–328.
Takahashi CD, Scheidt RA, Reinkensmeyer DJ. Impedance control and internal model formation when reaching in a randomly varying dynamical environment. J Neurophysiol 86: 1047–1051, 2001.
Tong C, Wolpert DM, Flanagan JR. Kinematics and dynamics are not represented independently in motor working memory: evidence from an interference study. J Neurosci 22: 1108–1113, 2002.
Wada Y, Kawabata Y, Kotosaka S, Yamamoto K, Kitazawa S, Kawato M. Acquisition and contextual switching of multiple nternal models for different viscous force fields. Neurosci Res 46: 319–331, 2003.[CrossRef][ISI][Medline]
Wigmore V, Tong C, Flanagan JR. Visuomotor rotations of varying size and direction compete for a single internal model in motor working memory. J Exp Psychol Hum Percept Perform 28: 447–457, 2002.[CrossRef][ISI][Medline]
Wolpert DM, Ghahramani Z, Jordan MI. An internal model for sensorimotor integration. Science 269: 1880–1882, 1995.
Wolpert DM, Kawato M. Multiple paired forward and inverse models for motor control. Neural Netw 11: 1317–1329, 1998.[CrossRef][ISI][Medline]
This article has been cited by other articles:
|
|
 |
|
  L. E. Brown, E. T. Wilson, M. A. Goodale, and P. L. Gribble Motor Force Field Learning Influences Visual Processing of Target Motion J. Neurosci., September 12, 2007; 27(37): 9975 - 9983. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
