Real-time error detection but not error correction drives automatic visuomotor adaptation

Abstract

We investigated the role of visual feedback of task performance in visuomotor adaptation. Participants produced novel two degrees of freedom movements (elbow flexion-extension, forearm pronation-supination) to move a cursor towards visual targets. Following trials with no rotation, participants were exposed to a 60° visuomotor rotation, before returning to the non-rotated condition. A colour cue on each trial permitted identification of the rotated/non-rotated contexts. Participants could not see their arm but received continuous and concurrent visual feedback (CF) of a cursor representing limb position or post-trial visual feedback (PF) representing the movement trajectory. Separate groups of participants who received CF were instructed that online modifications of their movements either were, or were not, permissible as a means of improving performance. Feedforward-mediated performance improvements occurred for both CF and PF groups in the rotated environment. Furthermore, for CF participants this adaptation occurred regardless of whether feedback modifications of motor commands were permissible. Upon re-exposure to the non-rotated environment participants in the CF, but not PF, groups exhibited post-training aftereffects, manifested as greater angular deviations from a straight initial trajectory, with respect to the pre-rotation trials. Accordingly, the nature of the performance improvements that occurred was dependent upon the timing of the visual feedback of task performance. Continuous visual feedback of task performance during task execution appears critical in realising automatic visuomotor adaptation through a recalibration of the visuomotor mapping that transforms visual inputs into appropriate motor commands.[PUBLICATION ABSTRACT]

Exp Brain Res (2010) 201:191207 DOI 10.1007/s00221-009-2025-9

RESEARCH ARTICLE

Real-time error detection but not error correction drives automatic visuomotor adaptation

Mark R. Hinder Stephan Riek James R. Tresilian

Aymar de Rugy Richard G. Carson

Received: 11 November 2008 / Accepted: 18 September 2009 / Published online: 15 October 2009 Springer-Verlag 2009

Abstract We investigated the role of visual feedback of task performance in visuomotor adaptation. Participants produced novel two degrees of freedom movements (elbow exionextension, forearm pronationsupination) to move a cursor towards visual targets. Following trials with no rotation, participants were exposed to a 60 visuomotor rotation, before returning to the non-rotated condition. A colour cue on each trial permitted identication of the rotated/non-rotated contexts. Participants could not see their arm but received continuous and concurrent visual feedback (CF) of a cursor representing limb position or post-trial visual feedback (PF) representing the movement trajectory. Separate groups of participants who received CF were instructed that online modications of their movements either were, or were not, permissible as a means of improving performance. Feedforward-mediated performance improvements occurred for both CF and PF groups in the rotated environment. Furthermore, for CF participants this adaptation occurred

regardless of whether feedback modications of motor commands were permissible. Upon re-exposure to the non-rotated environment participants in the CF, but not PF, groups exhibited post-training aftereffects, manifested as greater angular deviations from a straight initial trajectory, with respect to the pre-rotation trials. Accordingly, the nature of the performance improvements that occurred was dependent upon the timing of the visual feedback of task performance. Continuous visual feedback of task performance during task execution appears critical in realising automatic visuomotor adaptation through a recalibration of the visuomotor mapping that transforms visual inputs into appropriate motor commands.

Keywords Visuomotor adaptation Visual feedback

Sensory information Motor learning

Contextual (colour) cues

Introduction

The human motor system has a well-documented capacity to adapt to environments that are unlike any previously experienced. That is, behavioural changes (adaptations) occur such that, with practice, performance in the novel environment will resemble that observed in the usual environment. This adaptation may be mediated by a variety of different neural and cognitive mechanisms. However, for any adaptation to occur a discrepancy between the task goal and the actual movement outcome (an error) must rst be detected. The visual and somatosensory systems are the most important sources of such information and normally both systems are likely to contribute to error detection and, consequently, adaptation. In the case of mechanical perturbations, such as those imposed by torque motors (Shadmehr and Mussa-Ivaldi

M. R. Hinder S. Riek J. R. Tresilian A. de Rugy

R. G. CarsonPerception and Motor Systems Laboratory,School of Human Movement Studies,University of Queensland, Brisbane 4072, Australia

M. R. Hinder (&)

Motor Control Laboratory, School of Psychology, University of Tasmania, Private Bag 30, Tasmania 7001, Australiae-mail: [email protected]

J. R. TresilianDepartment of Psychology, University of Warwick, Coventry CV4 7AL, UK

R. G. CarsonSchool of Psychology, Queens University Belfast, Belfast BT7 1NN, Northern Ireland, UK

123

192 Exp Brain Res (2010) 201:191207

1994; Caithness et al. 2004), errors occur because the limb is perturbed forcefully from its intended trajectory. This (trajectory) error can, in most situations, be detected by both the visual and proprioceptive systems. When this is the case, it has been suggested that the brain weights the estimate from each modality according to the preciseness or reliability of each estimate (Ernst and Banks 2002; van Beers et al. 1999, 2002). However, vision only (Ghez et al. 1995) or proprioceptive information only (Lackner and DiZio 1994; Tong et al. 2002; Scheidt et al. 2005; Franklin et al. 2008) can be sufcient for adaptation to occur. In the case of altered visuomotor environments, such as those imposed by prism goggles (e.g. Lackner 1974) or computer generated transformations (Flanagan and Rao 1995; Wolpert et al. 1995; Ghahramani and Wolpert 1997) or rotations (Cunningham 1989; Krakauer et al. 1999; Wigmore et al. 2002), error is not induced by a perturbation of the limb from its intended trajectoryparticipants plan and execute a movement as intended. Rather, the error signal occurs due to discordance between where participants sense they have moved to (via proprioceptive feedback), and where the manipulated (by way of computer-mediated transformation, for example) visual feedback indicates they have moved. Accordingly in these circumstances, the visual feedback of task performance is necessary to drive visuomotor adaptation. Indeed, in the absence of this visual feedback (and any other feedback of task performance relative to the intended goal, for example tactile information) participants would simply have no means to detect any performance error and, as such, there would be no impetus for adaptation.

Exposure to a novel (e.g. rotated) visual environment is usually characterised by initial large errors followed by incremental improvements in task performance. The latter phase is referred to as visuomotor adaptation. This adaptation is thought to occur as a result of an automatic recalibration of the original (i.e. appropriate for the non-rotated environment) internal model (Kawato 1999) that characterises the relationship between the coordinates of the visual environment (inputs) and the issued motor commands (outputs), such that it represents the altered visual environment (Cunningham 1989). This recalibration serves to reduce, and eventually remove, the discordance (error) between proprioceptive and visually derived feedback. Subsequent removal of the rotated environment typically leads to aftereffects (Krakauer et al. 1999; Wig-more et al. 2002), whereby trajectory errors in the opposite direction to those observed upon initial exposure to the rotated environment are observed. Aftereffects are indicative of motor commands that are generated on the basis of the newly acquired (i.e. recalibrated) visuomotor mapping. Re-adaptation to the non-rotated task (a reversal of the recalibration) is required for performance in the non-rotated task to return to baseline.

We previously reported that in an isometric task (Hinder et al. 2008) adaptation to a visuomotor rotation occurred in a manner that gave rise to the subsequent expression of aftereffects only when visual feedback of performance was provided during task execution, i.e., when the discordance between proprioceptive feedback and visual feedback could be detected as the action was unfolding. When visual feedback was provided following task execution, the rate and extent of the visuomotor adaptations were similar to those observed when visual feedback was provided during task execution. In these circumstances, however, aftereffects were not observed. The possibility that in the latter instance a cognitively mediated strategy was used to compensate for the imposed rotation was supported by the observation that reaction times were elevated during the adaptation phase.

It is conceivable, however, that the paucity of movement-related proprioceptive feedback that is an intrinsic characteristic of the isometric task resulted in forms of adaptation that were abnormally dependent on the nature of the visual feedback of task performance that was provided. To investigate this possibility, in the present study we employed a dynamic visuomotor adaptation task which recreated as closely as possible the isometric task (Hinder et al. 2008). Participants made two degree of freedom movements of the elbowforearm complex in order to acquire visual targets. By its nature this task provides participants with proprioceptive feedback relating to changes of muscle force and limb position (muscle length), roughly analogous to that proprioceptive feedback available in everyday reaching tasks. In separate experimental conditions, visual feedback of task performance was available either during the movements (concurrently and continuously) or was provided following completion of each movement (terminal and discrete). It is apparent that provision of visual feedback during execution of a task permits online (or closed loop) error detection such that of motor commands can be modied to improve performance(i.e., adaptation). It is worth noting, however, that perception of the error is not necessary for adaptation (Pisella et al. 2000; Binsted et al. 2007; Klassen et al. 2005) Such online modications are not possible when visual feedback of performance is provided following task execution, in which case the adaptation is necessarily based on a representation of the movement outcome. To determine if the implementation of such online modications is a critical factor in determining the nature of visuomotor adaptation, we utilised two distinct concurrent (online) feedback conditions. In one, online modications were permissible; in the other they were not. We hypothesised that both online and post-trial visual feedback of performance would enable performance improvements in the rotated environment, consistent with our previous work (Hinder et al. 2008).

123

Exp Brain Res (2010) 201:191207 193

However, we wished to test whether the manner in which adaptation occurred varied as a function of the type (online or post-trial) of visual feedback. An adaptation mediated by a recalibration of the visuomotor map would be expected to elicit aftereffects following training, while adaptation by way of a cognitive strategy would not be expected to elicit aftereffects. These ideas were tested using specic (pre-planned) statistical comparisons based on angular errors of movement trajectories.

Methods

Participants

Twenty-four self-reported right-handed participants (23 37 years) took part in this study. All participants gave informed consent to the procedures, which were approved by the Medical Ethics Committee of the University of Queensland, and conformed to the Declaration of Helsinki.

Apparatus

Participants sat in a height-adjustable chair 1 m from a computer screen, positioned at eye level. The dominant right arm was placed in a manipulandum that allowed movement in two degrees of freedom (d.f.): elbow exion/extension

and forearm pronation/supination. The back of upper arm was placed against a vertical padded brace and held stationary, adjacent to the torso, by a Velcro strap. Participants grasped an instrumented vertical cylindrical handle, with pads positioned above and below in order to minimise movement of the hand. The chair height was adjusted for each participant such that the upper arm was aligned vertically and the centre of rotation of the elbow and the centre of rotation of the exion/extension axis of the manipulandum were coincident (Fig. 1; see Shemmell et al. 2005 for further details). This task represented the dynamic equivalent of the isometric task reported in our previous work (Hinder et al. 2008). As such, any change in the manner of the observed adaptation could be assumed to be as a result of a change in the nature of the proprioceptive feedback.

Potentiometers coupled to each axis of the manipulandum enabled real-time feedback of the angular position of the limb (exion/extension and pronation/supination). These kinematic recordings were sampled at 2,000 Hz at an analogue to digital interface (AT-mio-16E-10, National Instruments, TX, USA) and stored on a personal computer for further ofine analysis. Labview (vers 5.0, National Instruments, TX, USA) was used to write experimental control and data acquisition programs.

A vertical cloth screen was carefully positioned between the participants torso and arm, such that they were unable see their arm, or the manipulandum itself, throughout the

Fig. 1 Experimental set-up. a Apparatus. Participants held the manipulandum and controlled a visual cursor displayed on the screen via exionextension (updown cursor movement) and pronation supination (leftright cursor movement) movements of the elbow forearm complex (see b and c, respectively). Torque was applied to the servo-controlled motor controlling the exionextension axis such

that when participants relaxed their arm (with the upper arm against the restraint), their forearm was horizontal and the cursor appeared at the start position in the centre of the screen (depicted as a white dot). One of eight visual targets (represented as grey dots), equally spaced around the start position, was presented on each trial

123

194 Exp Brain Res (2010) 201:191207

experiment. Consequently, participants had to rely upon the visual feedback provided on the computer screen and proprioceptive/kinaesthetic feedback. By perturbing the visual feedback by way of a visuomotor rotation (see Experimental paradigm), we were able to introduce discordance between the position of the limb as indicated by the proprioceptive and visual sensory modalities.

Task procedure

A yellow dot at the centre of the screen, representing the start zone (zero displacement), was presented at the beginning of each trial. A white cursor indicated the current location of arm (in each degree of freedom). A gravity-compensation torque was applied via a servo-controlled motor (Brushless Servo Motor; Baldor Electric Co.) in the exionextension degree of freedom, to compensate for the gravitational effects corresponding to the combined weight of the manipulandum and the participants arm.1 As such, when each participant relaxed their arm (with the upper arm secured against the padded restraint, their forearm was horizontal and the cursor appeared at the start position in the centre of the screen (Fig. 1). In the pre-movement period (i.e., before presentation of the target in each trial) the white cursor representing current arm position was only visible when the arm remained within the start zone (forearm horizontal; palm vertical, i.e., midway between pronation and supination; 3.6). The visual feedback around the start position allowed participants to make small positional adjustments to ensure they were in the start position, but precluded them from investigating the arm movementcursor movement relationship over the wider workspace. The visual display (active screen area) was an 18 cm square region, on which the visual cursor moved1.5 cm for every 10 of joint rotation in either d.f. As such, 60 of pronation/supination or exion/extension from the central start position would result in the cursor being displaced to the limits of the visual display.

Each trial began with a period of random duration (12 s) during which participants were instructed to stay relaxed with the cursor in the start zone. One of eight targets, equally spaced at 45 intervals around, and at a constant distance of5.4 cm from the start zone, was presented, accompanied by an auditory tone. Each target represented a resultant joint rotation of 36. For single d.f. targets, a rotation of 36 in a single direction was required. For targets representing rotations in both d.f., the vertical (exion/extension) and horizontal (pronation/supination) requirements of the target were determined as the cosine of the subtended angle

between the target direction and the single degree of freedom target (45), multiplied by the joint rotation/target displacement for a single degree of freedom target. Accordingly, the 2 d.f. targets required 18 rotation in each degree of freedom. Participants were instructed to react quickly upon presentation of the target, and move the cursor towards the target as quickly as possible by moving their arm in one or both d.f.

The servo-control motors applied forces in proportion to the angular displacement in each degree of freedom (i.e., stiffness forces with an equilibrium position that coincided with the start position for each trial). For the vertical (exionextension d.f., this torque was imposed over-and-above the gravity-compensation torque described above). The magnitude of the resistive force (controlled by K in the equation s = -Kh, where s is the imposed torque and h is the angular displacement of the limb) did not vary between participants and experimental blocks (see Experimental paradigm). Pilot testing had revealed that participants could acquire the targets in each trial without considerable physical effort. Participants did not report the experiment to be physically fatiguing.

Experimental groups

Participants were randomly assigned to one of three groups (n = 8 per group). These groups undertook the experiment with varying types of visual feedback of task performance. Two groups were presented with continuous visual feedback of the cursor position over the whole workspace during the trial (i.e., concurrent feedback, CF), while the third group was provided with post-trial feedback (PF) following task execution (Table 1).

The PF group and one of the CF groups were instructed to react quickly upon presentation of each target and generate a single fast, un-corrected, arm movement to move the cursor from the start position toward the target. Participants subsequently relaxed such that their arm and the cursor returned to the starting position (elbow at 90;

forearm in a neutral position) in preparation for the next trial. Thus, for these groups, we assessed task performance in the absence of intended online modications.

The PF group were unable to see the cursor representing their position with respect to the visual target during each movement. Rather, they were presented with knowledge of performance (KP) depicting the cursor path, from movement onset (see Data Reduction and Analysis) to nal position following task execution. As such, this group will be referred to as the PFKP group. Final position(i.e., movement offset) was dened as the screen position corresponding to the exionextension (y-component) and pronationsupination (x-component) components of the maximum resultant joint rotation. Following the

1 This torque was determined for each participant on an individual basis to account for differences in the weight and length of participants arms.

123

Exp Brain Res (2010) 201:191207 195

Table 1 Task instructions and visual feedback conditions for the three participant groups

Group Paradigm/number of trials Visual feedback Task

CFFF PRE (black screen): 10 per target = 80 Continuous, concurrent

React quickly and move cursor towards targetdo not make corrections

ROT (blue screen): 40 per target = 320

POST (black screen): 10 per target = 80

480 total trials

CF FB

PRE (black screen): 10 per target = 80 Continuous, concurrent

React quickly and make any necessary corrections to acquire target

ROT (blue screen): 40 per target = 320

POST (black screen): 10 per target = 80

480 total trials

PFKP PRE (black screen): 10 per target = 80 Post-trial feedbackof cursor path (KP)

React quickly such that post-trial feedback shows path terminating near target- do not make corrections

ROT (blue screen): 40 per target = 320

POST (black screen): 10 per target = 80

480 total trials

movement, PFKP participants relaxed, such that their arm returned to the start position, where they waited for the feedback of their performance, which was presented 5 s after the trial began for a period of 3 s. Following this feedback period, the next trial began. Therefore, there was an 8 s interval between the onsets of the fore-periods in consecutive trials.

Participants in the rst CF group were instructed not to make any online (feedback mediated) corrections to their movements despite being able to see the cursor representing arm position throughout the task. Accordingly, this group will be referred to as the continuous feedback, feedforward control group (CFFF). For this group, the fore-period of consecutive trials began at 8 s intervals, as per the PFKP group.

The second CF group were instructed to react quickly when the target was presented, move towards and make whatever online corrections necessary in order to ensure that the target was acquired in each trial (i.e. it was a requirement for success that the cursor representing the magnitude and direction of movement in each d.f. in each trial coincided with the target zone for at least 100 ms). This variant of the task promoted online (visual feedback mediated) corrections. Therefore, this group is referred to as the concurrent visual feedback with feedback correction group (CFFB). The purpose of this group was to determine if executing feedback modications facilitated feed-forward adaptation compared to the condition where concurrent visual information was provided, but feedback modications were not promoted.

For the CFFB group, when the cursor was held within the target zone, dened as a region 5% around the specic visual target (representing a resultant joint displacement of 36 from the starting position, 1.8) for 100 ms a second auditory tone sounded, at which point participants relaxed and their arm (and cursor) returned to the start position. For this group, movement offset was dened as

the point 100 ms before the second tone. Pilot testing revealed that to provide CFFB participants with sufcient time to make online corrections to acquire the target a slightly longer trial duration was required. This was especially true in the rst few trials in the rotated task (see below for Experimental paradigm). Consecutive trials began at 9 s intervals. Although the rate at which trials were conducted in the CFFB group was slightly slower than the other two groups, the longer movement times for the CFFB group ensured that the time periods between completion of the task in one trial and commencement of the subsequent trial was very similar across all participant groups.

Experimental paradigm

All groups undertook a short practice block of trials (four trials per target), in which online feedback of the cursor position was available. The relationship between movement of the forearm and movement of the cursor was non-rotated: elbow exion and extension movement resulted in vertically upward, and downward, cursor movement, respectively. Pronation and supination movement resulted in cursor movement to the left, and right, respectively. Within each group of 16 trials, two trials to each target were undertaken; the order of target presentation within this cycle was pseudo-randomised. The practice block consisted of two such cycles. Despite provision of online feedback in this short pre-experimental block, PFKP and CFFF participants were instructed to make one rapid, uncorrected cursor movement towards the targetas per the instructions for the subsequent experimental blocks (see below). CFFB participants made trajectory modications as required, in order to acquire the presented target. The purpose of this block was threefold. Firstly, it ensured that all participants understood how to produce movements in each orthogonal d.f. Secondly, it enabled participants to

123

196 Exp Brain Res (2010) 201:191207

appreciate the magnitude of the joint rotations (movements) required to move to the vicinity of the target. This would ensure that in subsequent blocks for the PFKP group the position of the cursor relating to maximum movement amplitude was not beyond the limits of the visual display. Moreover, it prevented all participants from making gross over-rotations in either degree of freedom such that the manipulandum approached its working range.2

Finally, these trials taught the CFFF group to resist any urge they had to produce corrective movements on the basis of the visual information that was available to them. These practice trials were not included in any statistical analysis.

All groups subsequently undertook three experimental blocks of trials. All trials in a given block were conducted in the presence of a specic relationship between the direction of movement and the direction of the ensuing cursor movement. The background colour of the display, during both the trial and in the post-trial feedback period (for the PFKP group), indicated the operative relationship between the direction of movement and the resulting displacement of the cursor. The PRE block (10 trials to each target; 5 cycles; 80 trials total) was undertaken without any rotation imposed on the elbow movementcursor movement mapping i.e., the conditions were identical to those experienced in the short practice session. PRE trials were conducted with a black background. Participants could therefore associate this arm movementcursor movement relationship with the contextual cue (screen colour).

Following the PRE block, participants were exposed to a block of trials in which the feedback of cursor position was rotated 60 clockwise (CW) with respect to the direction of elbow movement (ROT block). By convention, this rotation was assigned a negative direction (-60). This block consisted of 40 trials to each target (20 cycles, 320 trials total). The vertical and horizontal target requirements were re-calculated according to the imposed 60 CW rotation, such that the resultant joint rotation required to acquire each target remained at 36. All ROT trials were performed with a blue background throughout the trial and, for the PFKP group, during the post-trial feedback periods. Participants were thus provided with the facility to associate the novel arm movement-cursor movement relationship

with the blue screen. Finally, participants were re-exposed to the original (non-rotated) mapping which they had previously encountered in the PRE block (POST block). A black screen was used, as in the PRE block, allowing the restoration of the original (non-rotated) visuomotor relationship to be identied. The purpose of the POST block was to probe for aftereffects that may have occurred as a result of exposure to the rotation. The group-specic visual conditions, and task instructions, discussed above were relevant in each of the three experimental blocks for each group. The PFKP group received only post-trial feedback in each block). The presentation of targets within each cycle was pseudo-randomised such that each participant within a group was presented with a different target on the rst trial of each trial block.

In total, the experimental protocol lasted *75 min. There was a brief break (\1 min) between consecutive blocks of trials in which participants remained seated with their arm relaxed. All participants were explicitly reminded at the start of the each experimental block to react quickly upon presentation of each target, and that the relationship between their movement and movement of the cursor in all trials was represented by the colour of the screen. As such, participants had sufcient information to enable them to identify the non-rotated environment at the commencement of the POST block.

Data reduction and analysis

Movement time series, representing cursor position in the two orthogonal degrees of freedom (exionextension and pronationsupination of the forearmelbow complex), were digitally low-pass ltered at 15 Hz with a second-order, dual-pass Butterworth lter. Movement onset was determined using an algorithm based on cursor speed (Teasdale et al. 1993), with an initial threshold of 15% of the maximum tangential speed. For the CFFB group, movement offset was determined as the time 100 ms prior to the second tone, i.e., the start of the rst period of at least 100 ms duration in each trial in which subjects remained in the target zone. For the CFFF and PFKP groups, movement offset was determined as the point of maximum resultant joint rotation.

Movement time, MT, from movement onset to movement offset, i.e., target acquisition (CFFB group) or maximum resultant joint rotation (other groups), was used to assess performance within groups. This measure was not used as the basis of comparison between groups as extra time was required by CFFB participants to issue corrective motor commands to acquire the target. For the groups who were instructed not to make feedback modications, we predicted that MT would be unaffected by imposition of the rotation. For the CFFB group, however, longer MTs

2 As a precaution to help prevent this scenario, the servo-motors controlling movement in each degree of freedom were programmed with elastic walls at the limits of the visual display, i.e., very large torques were required to move beyond a 60 rotation from the start position. If the manipulandum approached these limits, participants would feel a strong spring-like force that repelled them away from region corresponding to the edge of the visual display. We note, however, that no participants hit the walls within the experimental trials.

123

Exp Brain Res (2010) 201:191207 197

would be predicted upon initial exposure to the rotation due to increased requirements for feedback modications to acquire the target. Reductions in MT during the course of each block (PRE, ROT, POST) would be expected for two reasons: rstly, progressively more accurate feedforward commands would result in smaller feedback corrections being required, and secondly with practice participants should be able to implement the required feedback corrections more quickly.

Angular error, h, was calculated at two points within the trial: 100 ms after movement onset, and at the peak resultant velocity (i.e., the peak resultant velocity of the arm). We calculated the average angle between the vector dened by the start and target positions, and the vector dened by the start position and actual cursor position, in a 10 ms window centred at either 100 ms after movement onset or peak resultant velocity, respectively. On the basis of the assumption that no responses based on the online visual feedback provided to the CFFB group could occur prior to 100 ms, this measure assessed feedforward performance. The angle at peak resultant velocity was calculated to take account of differences in movement times between participant groups (see Results) i.e., 100 ms following movement onset represented a different position in the trial, with regard to acceleration and deceleration phases. Hand path characteristics up to peak resultant velocity are generally accepted as being independent of online corrections, i.e., representative of feedforward performance.

We predicted that increased angular errors would occur for all groups due to imposition of the rotation. Reduction of angular error within the training (ROT) block would signify compensation for the rotation. Angular errors in the POST block that were in the opposite direction to those exhibited upon initial exposure to the rotation (aftereffects) would indicate that participants implemented motor commands more appropriate for the rotated context. However, accurate performance in the POST block would indicate that participants correctly implemented commands for the non-rotated task.

Reaction time, RT, was calculated as the interval from target presentation to movement onset, to assess the extent of pre-movement processing. Adaptation to the rotation through mechanisms that amended the visuomotor map(i.e., implicit adaptation, Mazzoni and Krakauer 2006) would not be expected to result in increased reaction times. If, however, participants adapted in a manner that required explicit decisions or compensation strategies to be implemented, planning time may increase and result in longer RT.

We determined cycle-average (over the 16 trials of each cycle; two trials to each target location) data for the ve cycles of the PRE and POST block, and the 20 cycles of the

ROT block for each participant group for each dependent variable (h, RT and MT). The values for the rst and last cycle of each movement block (PREintial, PREnal, ROTinitial, ROTnal, POSTinitial and POSTnal) were used to conduct statistical comparisons of performance (averaged across all target conditions) for each dependent variable. Assessing performance based on cycle-averages is common within the motor learning literature (e.g. Krakauer et al. 1999; Caithness et al. 2004) as it allows one to assess performance in a manner that is representative of movements to all targets, rather than just a single target.

A three-way (group {CFFB, CFFF, PFKP} 9 block {PRE, ROT, POST} 9 time {initial, nal}), ANOVA with repeated measures on the nal two factors was conducted for each dependent variable. However, in the context of the present experimental design, the majority of the main effect and interaction terms associated with the ANOVA are of no real benet in aiding with the interpretation of the results. The purpose of the ANOVA was to compute the values (i.e. sums of squares, etc.) that enabled us to undertake planned comparisons. Because the comparisons we report were all planned, they can be reported in the absence of signicant main effect or interaction terms (c.f. post hoc tests, where main or interaction effects must be signicant to allow post hoc analyses; see Keppel 1982). PREinitial was compared with PREnal to assess any performance changes within the non-rotated (pre-training) block, while comparing ROT-initial with ROTnal enabled us to determine if there was a signicant performance improvement within the rotated environment. POSTinitial was compared with PREnal to determine any changes in performance in the non-rotated condition as a result of exposure to the rotation block, i.e., to probe for aftereffects as a result of exposure to the rotated environment. This particular comparison provides a robust test for aftereffects, because it is independent of the degree of adaptation during training. A nal comparison of POSTinitial with POSTnal assessed whether there were any reliable changes in performance over the course of the POST block. Adjustments to alpha (modied Bonferroni adjustment) were made as necessary to account for multiple comparisons. Effect sizes, f, were calculated for each of the planned comparisons (Cohen 1969), to aid in the interpretation of the tests of signicance. The effect size describes the degree of departure from no effect, in other words, the degree to which the phenomenon is manifested.

The rate of performance improvements (reduction in angular error) in the rotated environment were compared across groups by tting a power function to the cycle-averaged trial data in the training (ROT) block for each participant, using a least-squares t criteria (Newell and Rosenbloom 1981). This function takes the form

123

198 Exp Brain Res (2010) 201:191207

y axbwhere y represents the angular error in cycle x, a indicates the (tted) level of performance in the rst cycle in the rotated environment and b represents the slope of the curve, and is an indication of the rate of adaptation. While tting curves to the cycle-average data is consistent with our use of cycle-averages to determine the initial and nal performance in the rotated context (c.f. Caithness et al. 2004; Krakauer et al. 1999), the curve parameters may not be sufciently sensitive to detect subtle differences in the adaptation rates which may apparent when considering individual trials. Accordingly, we also tted curves to the individual trial data for each participant (in this instance y represents the error in trial x). One-way ANOVA was used to compare adaptation rates (b parameter from the curve

tting procedure) for the three feedback groups, for each of the two tting methods independently.

Results

Figure 2 shows cursor trajectories in the PRE, ROT and POST blocks for a representative participant from each group. Each participant produced relatively straight paths in the PRE block (left panels). In the initial ROT trials (central panels; thin traces), all groups exhibited large angular errors early in the movements as a result of the imposed rotation. The CFFB group exhibited large hooks in the trajectory indicating online corrections such that the target was attained (Fig. 2a; thin traces). In contrast, the other groups exhibited angular errors at the

A

CF-FB

PF-KR

Fig. 2 Example cursor paths. Paths are shown for the rst (thin line) and nal (thick line) trial in each block (PRE, left; ROT, centre; POST, right) to the exion (top), extension (bottom), pronation (left) and supination (right) targets for the a CFFB b CFFF and c PF KR groups. For the CFFB group, online corrections ensured that all paths terminated at the desired target. However, terminal errors remained for the other groups as online corrections were not permitted (When required, these trajectories are labelled to ensure it is clear to which target the participant was aimingsee key). All groups exhibited performance improvements within the training (ROT) block; however, only the groups who received online visual feedback exhibited aftereffects when the rotation was removed (POST block). Data shown are individual trials of a single participant in each of the experimental groups

PRE POST

ROT

CF-FF

B

PRE POST

ROT

PRE POST

ROT

Ff

Pf Sf

Ef

Ff

Si

Pi

Ei Si

Fi

Fi

Sf

Pf

Ei

Pi

Ef

C

Ff

Si

Fi

Sf

y (cm)

Pf

Ei

Pi

Ef

5.4 0 5.4

x (cm)

FLEXION, E (up) SUPINATION, S (right)

EXTENSION, E (down) PRONATION, P (left)

Initial (i) trial: thin Final (f) trial: thick

123

Exp Brain Res (2010) 201:191207 199

movement endpoint similar to the errors observed earlier during the movement, suggesting that corrective movements were not made (Fig. 2b, c; thin traces). All groups had signicantly straightened hand paths in the initial part of each movement by completion of the ROT block (Fig. 2; central panels, bold traces). However, only the groups provided with continuous visual feedback exhibited aftereffects when the rotation was removed (Fig. 2a, b; right panels, thin traces); the post-trial feedback group manifested paths straight towards the targets (Fig. 2c; right panel, thin traces), resembling those observed immediately prior to the rotation.

To corroborate that the CFFF and PFKP groups adhered to the task instructions and did not make corrective movements we calculated the time derivative of individual movement proles. After applying a low-pass (6 Hz Butterworth) lter to remove noise associated with the differentiation process, we calculated the number of peaks in these velocity proles and determined the mean number of velocity peaks for each 16-trial cycle of the ROT block (Fig. 3). Two-way ANOVA revealed a signicant group effect (p \ 0.001, gp2 = 0.88) indicating that the number of velocity peaks varied between groups. Subsequent pair wise comparisons indicated that the CFFB group

exhibited a higher number of corrections than either of the other two groups (p \ 0.001 in both cases) but that the number of corrections in the PFKP and CFFF groups did not differ signicantly (p [ 0.99). The cycle-number main effect was also signicant (p \ 0.001, gp2 = 0.42) indicating that, averaged over the three groups, the number of corrective movements fell with training. However, the interaction between cycle-number and group was also signicant (p \ 0.001, gp2 = 0.59). This interaction, together with visual inspection of Fig. 3, clearly indicates that the cycle main effect was driven solely by the reduction in the number of velocity peaks across repeated cycles by the CFFB group. In contrast the other two groups had a low (*1) number of velocity peaks throughout the training period. This result indicates that both the CFFF and PF

KP groups adhered to instructions and did not make corrections to their movements during each trial.

Figure 4 shows angular error (measured at peak resultant velocity, see Data reduction and analysis and Angular error) in each trial of the PRE, ROT and POST blocks, averaged over the 8 participants in each groups. Figure 5ab shows cycle-averaged data (for angular errors and movement times) derived from the individual trial data in Fig. 5.

The cycle-average data presented in Fig. 5 indicate (qualitatively) that throughout the PRE block, when averaged over the eight targets, participants produced movements that resulted in relatively accurate movement of the cursor towards the target (Fig. 5a, PRE block cycles). The low magnitude of h throughout the PRE block, together with low variability of error between the cycles of the PRE block indicates that participants found the arm movement cursor movement mapping fairly intuitive (i.e., there is no trend in the angular errors within the PRE block to indicate that any adaptation occurred with respect to movement direction/control). We note that small angular errors were observed throughout the short practice block (data not shown), suggesting that the low errors in the PRE block were a result of participants nding the non-rotated visuomotor mapping intuitive, rather than a result of rapid adaptation in the practice block. Exposure to the 60 CW

4

CF-FB CF-FF PF-KP

Average velocity peaks

3

2

1

0 0 5 10 15 20Cycle number

Fig. 3 Average number of peaks in the velocity prole. Data shown is the group-average for each cycle in the ROT block (CFFB squares; CFFF triangles; PFKP diamond; also see key)

A B C

45

CF-FB CF-FF PF-KP

TRIAL/BLOCK TRIAL/BLOCK TRIAL/BLOCK

45

45

30

30

30

15

Angular Error ()

15

15

0

0

0

-15

-15

-15

-30

-30

-30

-45

-45

-45

-60

-60

-60

-75 0 80 160 240 320 400 480

-75 0 80 160 240 320 400 480

-75 0 80 160 240 320 400 480

PRE ROT POST

PRE ROT POST

PRE ROT POST

Fig. 4 Angular errors measured at peak joint velocity- individual trial data. Data shown is the group-average for each trial of the PRE, ROT and POST blocks for the a continuous feedback with online

corrections CFFB b continuous feedback with no online corrections CFFF and c post-trial feedback PFKP groups

123

200 Exp Brain Res (2010) 201:191207

CF-FB CF-FF PF-KP

CF-FB CF-FF PF-KP

A

45

C

45

30

30

* *

*

Angular error () MT (S)

MT (S) Angular error ()

15

15

*

*

0

0

-15

-15

-30

-30

-45

-45

-60

-60

-75

* * *

B

2.0

D

*

1.5

1.0 *

0.8

1.0

0.6

*

0.4

0.5

0.2

0.0 0 5 10 15 20 25 30

0.0

-0.2

ROT initial ROT final POST initial

PRE ROT POST

CYCLE/BLOCK

-75

CYCLE

Fig. 5 Cycle-average angular errors and movement times. Data is shown for the PRE, ROT and POST blocks, averaged across the eight participants in each group (CFFB squares; CFFF triangles; PFKP diamonds; also see key). a Angular error and b Movement time. Note that in b the data for the CFFB and CFFF groups are coincident in the POST block cycles (i.e., the triangles are predominantly hidden behind the squares). For all groups, the PRE and POST block data are presented as black symbols, while the ROT block data appears as grey

symbols. cd Mean (95% condence intervals) difference (D) in angular error (c) and movement time (d) in the ROTinitial, ROTnal and POSTinitial cycles, relative to the PREnal cycle. A asterisk above the ROTinitial or POSTinitial bars indicates a signicant change relative to the corresponding PREnal value (p \ 0.05).

Signicant changes between ROTinitial and ROTnal are depicted by horizontal lines and asterisk

rotation resulted in angular errors, observable in the direction of the rotation (i.e., negative errors by convention) for all groups (Fig. 5a, ROT block). The fact the magnitude of the errors were somewhat less than the imposed rotation of 60 indicates that a certain degree of adaptation occurred within the rst cycle of trials, a characteristic that can be observed in Fig. 4. With repeated exposure to the rotation, all groups were able to reduce the angular errors in the initial period of the movement, indicating modication of the feedforward motor commands (Fig. 5a, nal ROT cycles). Following completion of the training (ROT) block, performance with respect to angular error was similar across the three participant groups, and was of the order of -10 (i.e., participants did not quite compensate fully for the imposed rotation of -60). We note that the CFFB group, who were instructed to make corrections to movements within each trial, were able to accurately acquire the targets on the vast majority of trials,i.e., despite exhibiting angular errors at peak velocity comparable to the other groups, they reduced these errors in the deceleration phase of the movement, and located the target (to within 5% of the specied target joint rotations; see Methods)

For the both the CFFF and CFFB groups, re-exposure to the non-rotated environment (POST block) resulted in

angular errors in the opposite direction to those errors observed when they had rst encountered the rotation (Fig. 5a, initial POST block cycles, squares and triangles),i.e., aftereffects were observed. By the end of the POST block, paths had become signicantly straighter for both CF groups. In contrast, the PFKP group implemented motor commands (i.e., generated arm movements) in the initial POST block trials that resulted in cursor movements that were accurately directed towards the targets (Fig. 5a initial POST cycles, diamonds), i.e., no aftereffects were apparent. The presence of aftereffects for the continuous feedback groups, together with the absence of aftereffects for the post-trial feedback group is also clear in Fig. 4. Thus, although all groups appeared to be able to improve feedforward performance similarly in the rotated environment, only the CF groups performance in the non-rotated environment was affected as a consequence.

Figure 5cd depicts the angular errors and movement times in the ROTinitial, ROTnal and POSTinitial positions, relative to the value of the corresponding PREnal value. These panels allow the reader to clearly see the how each dependent variable varied within the training block, and upon re-exposure to the non-rotated task, with respect to the value of the variable following completion of the PRE block. Statistical analyses (pre-planned comparisons as indicated in Data

123

Exp Brain Res (2010) 201:191207 201

Reduction and Analysis) were performed on the cycle-average data for the rst and last cycle of each movement block for each dependent variable. These are reported below.

Angular error

We investigated feedforward performance by considering angular deviations of the cursor path at peak velocity, i.e., before any corrective movements would have occurred for the CFFB group. We note that the time at which peak velocity occurred (265328 ms following movement onset) did not vary signicantly across the three groups (one-way ANOVA, p [ 0.20; f = 0.10).

Initial exposure to the rotation (ROTinitial) led to an increase in h to relative to the PREnal cycle for all groups (p \ 0.05 for all groups; f = 0.77, 1.20, 0.97 for the CFFB,

CFFF and PFKP groups, respectively). By completion of the training (ROTnal), all groups performed more accurately than they had upon initial exposure to the rotated task context (p \ 0.05; f = 0.53, 0.99, 0.77 for CFFB, CFFF and PFKP, respectively). Aftereffects as a result of exposure to the rotated environment were assessed by comparing h in the cycle immediately prior to (PREnal) and following (POSTinitial) the rotation block. For both CF groups this comparison was statistically signicant (p \ 0.05) and was associated with a large effect size (f = 0.66, 0.62 for the

CFFB and CFFF groups), indicating the presence of aftereffects. In contrast, for the PFKP group this comparison was not statistically signicant (p = 0.49), was associated with a small effect size (f = 0.07) and suggests that no aftereffects were present for this group.

Inter-group comparisons of nal performance

A one-way ANOVA indicated that the angular error in the nal ROT cycle, (i.e., upon completion of the training period) did not differ signicantly across the three groups (p = 0.77; f = 0.16). Furthermore, we also determined that the co-efcient of variation of angular error (measured at peak velocity) within the 16 trials of the nal training (ROT) cycle did not vary as a function of group (One-way ANOVA, p = 0.38, f = 0.14). As such, following training, both performance and performance variability were comparable across groups. Accordingly, the presence of aftereffects for the CF groups, and absence of aftereffects for the PFKP group (reported above) cannot be attributed to inter-group differences in the level of skill acquired during the training epoch or the variability of performance.

Individual trial analysis

While cycle-averaging is a common technique enabling performance to be assessed across all target directions, it is

conceivable that such a process could mask subtle changes in performance that were present in the rst few trials of a cycle. Specically, we wished to ensure that cycle-averaging did not conceal any post-training aftereffects for the PFKP group which were quickly reduced (re-adaptation in the non-rotated environment) and as such not apparent in the POSTinitial cycle.

Linear regression was conducted on the 16 individual trials of the POSTinitial cycle. For the PFKP group, the gradient of the regression was not signicantly different to zero (p = 0.83), indicating there was no signicant trend in angular error within the 16 trials of the rst cycle in the POST block. Furthermore, the intercept parameter of the regression was not signicantly different to zero (p = 0.16). Indeed, the PFKP group-average angular error in the very rst POST trial was 1.3. This suggests that aftereffects were not manifested by this group, even in the very early POST block trials. For both CF groups the gradient of the linear regression was less than zero, although this did not reach signicance for the CFFB group (p = 0.10 for CFFB; p \ 0.05 for CFFF).

Moreover, the intercept parameters of the regressions for these groups were both signicantly greater than zero (p \ 0.05), indicating signicant aftereffects. These results suggest the substantial aftereffects exhibited by the

CF groups were incrementally reduced (to some degree) over the course of the 16 trials of the rst cycle (as supported by Fig. 4).

Overall, the angular error results suggest that the ability to issue appropriate motor commands when re-exposed to the non-rotated environment following exposure to an altered visual environment varied as a function of the visual feedback of task performance (concurrent or post-trial) that was provided.

Comparisons of adaptation rates

Power ts were derived from the angular error data using both cycle-average data and individual trials (see Methods). In all cases, the ts were statistically reliable (p \ 0.05) such that the parameters derived could be assumed to be a true representation of the dynamics of the adaptation process.

Cycle-average ts

Average r2 values for these ts were 0.37, 0.65 and 0.66 for PFKP; CFFF; and for CFFB groups, respectively. One-way ANOVA failed to show any signicant differences in the adaptation rates across the three groups (p = 0.51; b parameter = -0.47, -0.61, -0.55 for the PFKP, CFFF and CFFB groups, respectively) suggesting that despite differences in the visual feedback and task

123

202 Exp Brain Res (2010) 201:191207

instructions, participants were able to reduce error at the same rate within the training period.

Individual trial ts

To assess subtle differences in the adaptation rates that may have been obscured as a result of cycle-averaging, we tted power functions to the individual trial data for each individual subject. While the r2 values (average 0.10, 0.18 and 0.24 for PFKP, CFFF and CFFB groups, respectively) were lower than r2 values for those ts derived from cycle-average data, the ts were still reliable (p \ 0.05). One-way ANOVA indicated that, in contrast to the cycle-average analysis, there was a statistically reliable difference in the adaptation rates (b parameter) between the three groups (p \ 0.005; b = -0.10, -0.17, -0.18 for the PFKP, CFFF and CFFB groups, respectively). Subsequent post hoc analyses indicated that the adaptation rates for the two groups provided with continuous visual feedback were not dissimilar (p = 0.66), but that the post-trial feedback group appeared to adapt more slowly than both continuous feedback groups (p \ 0.05).

Movement time: a measure of overall task performance

MTs varied little between and within blocks for the CFFF and PFKP groups (Fig. 5b), with all of the pre-planned comparisons yielding non-reliable outcomes (p [ 0.20;

f \ 0.12 for all comparisons). Cycle-average MTs ranged between 0.50 and 0.66 across all block for these groups.

The similarity of MTs for the CFFF and PFKP groups suggests that CFFF participants adhered to instructions and produced fast movements, with no online corrections (despite the provision of online feedback).

CFFB participants were instructed to make whatever adjustments to the movement/cursor were necessary to

ensure that the target was acquired. The MTs exhibited by this group were markedly longer than those generated by the other participant groups in all cycles (Fig. 5b). Because we found the time at which peak velocity occurred was similar across all groups, the longer movement time for the CFFB group were principally a result of an extended deceleration phase in which the online feedback was used to correct movement errors, i.e., the longer movement times are indicative of the implementation of feedback control. The use of this mode of control is likely to account for the increased MTs, with respect to the PREnal cycle, exhibited by this group when they were initially exposed to the rotation (ROTintial) (p \ 0.05; f = 1.05). MT fell signicantly within the training period (ROTinitial vs.

ROTnal: p \ 0.05, f = 0.79), most likely reecting reduced dependence upon feedback control as the feed-forward commands became more specic to the rotated environment. Re-exposure to the non-rotated task (POST-initial) resulted in movements that were signicantly longer than those generated in the non-rotated cycle immediately prior to the rotation block i.e., PREnal cycle (p \ 0.05, f = 0.79). This increase in MT is consistent with a requirement to implement feedback corrections in order to compensate for the angular deviations (i.e., aftereffects) generated in the feedforward segment of the movement.

Reaction time (RT)

Reaction time from presentation of each target to movement initiation in each trial is shown in Fig. 6. Exposure to the rotation led to increased RTs in the ROTintial cycle for the PFKP and CFFF groups, with respect to RTs at completion of the PRE block (PREnal vs. ROTintial: p \ 0.05; f = 0.54, 0.92 for the CFFF and PFKP groups,

CF-FB CF-FF PF-KP

CF-FB CF-FF PF-KP

A B

1.0

0.6

* *

*

0.4

RT (S)

0.5

RT (S)

*

0.2

0.0

0.0

PRE ROT POST

BLOCK

ROT initial ROT final POST initial

CYCLE

Fig. 6 Cycle-average reaction time. a Cycle-averaged RT in the PRE, ROT and POST blocks, averaged across the eight participants in each group (CFFB squares; CFFF triangles; PFKP diamonds; also see key). b Mean (95% condence intervals) difference (D) in RT in the ROTinitial, ROTnal and POSTinitial cycles, relative to the

PREnal cycle. A asterisk above the ROTinitial or POSTinitial bars indicates a signicant change relative to the corresponding PREnal value (p \ 0.05). Signicant changes between ROTinitial and

ROTnal are depicted by horizontal lines and asterisk

123

Exp Brain Res (2010) 201:191207 203

respectively; Fig. 6a, b). For the CFFB group, this comparison did not reach statistical signicance (p = 0.13; f = 0.20). For the CF groups, RT decreased within the ROT block, although this result did not quite reach statistical signicance for the CFFB group (p = 0.07; f = 0.24 for the CFFB; p \ 0.05, f = 0.37 for the CFFF group).

In contrast, RT increased for the PFKP group within the ROT block (p \ 0.05; f = 0.27). For the CF groups, RT in the POSTinitial cycle were not dissimilar to those prior to adaptation to the rotation (PREnal) (p = 0.67, 0.60; f = 0.06, 0.13 for the CFFB, CFFF groups, respectively). For the PFKP group, despite a substantial fall in RT when returning to the non-rotated task, POSTinitial RT was still somewhat elevated with respect to the PREnal RT (p \ 0.05; f = 0.42). RT did, however, continue to fall within the POST group to levels similar to that seen in the

PRE block.

Discussion

When a motor command is generated in a novel environment the predicted sensory consequences of that motor command may not correspond with the actual sensory feedback received as a result of that command. Detection of this sensory error is critical to allow motor adaptation to occur. In the current visuomotor adaptation paradigm, the putative error signal was the discordance between task performance as indicated by visual feedback (indicating where participants moved to in relation to the intended target), and participants proprioception based estimate of the location to which they moved. While proprioceptive information regarding the movement is clearly present during the execution of the task, we manipulated the timing of the visual feedback of task performance, such that it occurred during or following task execution. In this manner we were able to investigate whether temporal separation between feedback from the proprioceptive and visual systems (resulting in a delay between task execution and error detection) affected the extent and/or nature of feedforward adaptation to a novel visuomotor environment in this goal-directed two degree of freedom movement. This experiment represents a logical progression from previous work conducted in our laboratory (Hinder et al. 2008) that investigated a similar research question using an isometric paradigm. In the previously reported case there was, by denition, no movement of the limb. In neither study was there direct visual feedback of the task-relevant state of the limb (joint torque or position). Accordingly, the only difference between the isometric and dynamic tasks in regard to sensory feedback was the additional proprioceptive feedback resulting from movement of the limb that was available in the dynamic task.

Previous work in our laboratory (Shemmell et al. 2005) indicated that initial performance and the subsequent extent of adaptation in this two degree of freedom movement task varied as a function of target position, even in the absence of a visuomotor rotation. The magnitude of the inter-target differences reported by Shemmell and colleagues were, however, very small compared to those that are induced by the visuomotor rotation in the present study. Moreover, in the present study movements to on-axis (i.e., horizontal/ vertical visual targets) and off-axis (i.e., diagonal visual targets) targets both required rotations of the forearm joints in both degrees of freedom (i.e., a combination of pronation/supination and exion/extension) when exposed to the 60 rotation. For these reasons, a comparison of inter-target differences in performance and/or comparisons of targets requiring movements in one or two degrees of freedom is not conducted here. Instead we focus on results from target-averaged data.

By completion of the training (ROT) period, groups provided with continuous (concurrent) visual feedback of performance (CFFF, CFFB) or post-trial visual feedback of performance (PFKP) had all attained a similar level of performance, compensating for the 60 CW rotation by about 50 (i.e., cursor paths following training still displayed about 10 of error; Fig. 6a, c). This under-compensation was also observed in our previous study in which an isometric task was examined (Hinder et al. 2008) and may be partially explained by anecdotal reports of some participants that they believed the rotation to be 45 (i.e., a rotation equivalent to one target position on the visual display). Residual angular errors of this magnitude are very similar, or indeed smaller, than those that persisted following the adaptation phase in a number of previous visuomotor rotation studies (e.g., Miall et al. 2004, Wang and Sainburg 2004, Krakauer et al. 2000, 2005, Hinder et al. 2007).

To determine if the action of making feedback corrections on the basis of the detected error signal promoted more rapid or more complete adaptation, or indeed changed the manner in which the adaptations occurred, we compared adaptation for participants provided with continuous visual feedback who were instructed not to make corrections to their movements, with that exhibited by participants provided with continuous visual feedback who were permitted to make corrections. We observed that the occurrence of corrections was not associated with feed-forward adaptation that was any faster or more complete than that observed when only feedforward control was permitted. This nding replicates results recently reported by Tseng et al. (2007), indicating that detection of an error signal allows feedforward adaptation to a novel visuomotor environment but concurrent corrections on the basis of the detected sensory error do not necessarily facilitate this adaptation.

123

204 Exp Brain Res (2010) 201:191207

A fundamental feature of our experimental design was the use of contextual (colour) cues to enable the rotated and non-rotated task contexts to be identied. As such, participants were provided with the information required to issue the appropriate motor commands upon re-exposure to the non-rotated task, if indeed these commands were still available following learning in the rotated environment. Using this method, we were able to expand upon Tseng et al.s work and investigate more closely the nature of the adaptation that occurred, with and without online visual feedback, and when online corrections were or were not permitted. Upon re-exposure to the non-rotated task (identied by the cue) signicant aftereffects were observed for both CF groups, but not for PFKP group. This nding is consistent with our previous work (Hinder et al. 2008) and conrms that that timing of the visual feedback of task performance (allowing a sensory error to be detected) has a profound effect on the nature of the feedforward adaptation that occurs in response to a novel visuomotor rotation. The results obtained when visual feedback of task performance was available during task execution are consistent with the established view that visuomotor adaptation occurs through an automatic modication of the visuomotor map that transforms sensory information into motor commands (Cunningham 1989; Krakauer et al. 1999). Furthermore, the nature of this adaptation was not affected by the absence or presence of online corrective actions on the basis of the online feedback.

We note that when rst challenged with the rotation, reaction time (RT) increased to some extent for both CF groups (although this was only statistically reliable for the CFFF group, Fig. 6). This increase in reaction time may be indicative of greater cognitive load (Klapp 1995) in the pre-movement (planning) period. With further exposure to the task, however, RT fell for CF groups to levels not dissimilar to those observed in the initial (pre-rotation) block, suggesting that execution of the task in the rotated context became more automatic. For these groups, performance improvements in the novel rotation were followed by signicant aftereffects upon re-exposure to the non-rotated context. The fact that aftereffects occurred despite the provision of a salient cue that identied the change in task environment (consistent with the results of Miall et al. 2004) lends weight to the proposition that aftereffects were due to an automatic recalibration of the visuomotor map, and not due to misidentication of the task context. The re-adaptation that occurred in the post-training block is consistent with a re-modication of the visuomotor mapping to characterise the non-rotated task context.

In contrast, the results obtained from those participants who received post-trial visual feedback suggest that the

nature of their adaptation was quite distinct. Despite receiving feedback of the whole trajectory (i.e., the PF participants received the same amount of visual feedback) the fact that the information was provided following movement completion appears to have led to quite a different adaptation process. The data are consistent with the proposition that, when challenged with the rotation, these participants implemented an explicit (cognitive) strategy that enabled performance improvements to occur to the same extent as those subjects who were provided with continuous visual feedback. Consistent with this proposition, substantially longer reaction times were exhibited throughout the ROT block (Fig. 6). Over the course of the block, RT actually increased slightly, suggesting that the cognitive demand did not fall, and the level of automaticity achieved with CF did not occur with PF. In principle, an effective strategy would be to employ a visuomotor mapping appropriate for the original task environment, in conjunction with an offset of the visual coordinates of the target. Once the visual input (target location) had been shifted (or offset) by the required amount (direction and magnitude) the original visuomotor mapping could be used to transform this information into appropriate motor commands. A conceptual model of the putative modes of adaptation for the PF and CF groups is shown in Fig. 7. In this scheme, the (putative) cognitive component required by the PF group may account for the observed increase in RT.

For the post-trial feedback group re-exposure to the non-rotated context resulted in accurate performance (no aftereffects) and a substantial reduction in reaction time. This nding is also consistent with the proposed strategy-based model, insofar as a salient contextual cue (in this case the colour of the background screen) identifying the non-rotated task permitted these participants to immediately adopt an approach to the task that was indistinguishable from that which had been employed prior to the imposition of the rotation (i.e., they could simply abandon the cognitive component of the model shown schematically in Fig. 7b).

With a view to providing a quantitative means to assess the rate at which performance improvements occurred in the altered visual environments, we tted power curves to the angular error data. In the rst instance we chose to t the curves to the cycle-average data, consistent with our analysis of the initial and nal performance in each trial block in which we also employed cycle-average measures (also see Krakauer et al. 1999; Caithness et al. 2004). The b parameter provides a measure of the rate of performance improvement, and it did not differ between the three groups. This evidence suggests that the participants adapted to the imposed rotation at the same rate irrespective of the nature of the visual feedback (concurrent or post-trial)

123

Exp Brain Res (2010) 201:191207 205

A B

NEW VISUOMOTOR TRANSFORMATION

VISUAL INPUT MOTOR OUTPUT

AUTOMATIC

VISUAL INPUT

MOTOR OUTPUT

OFFSET TARGET

ORIGINAL VISUOMOTOR TRANSFORMATION

COGNITIVE AUTOMATIC

Fig. 7 Conceptual models for visuomotor adaptation. a Visual feedback of task performance during task execution permits adaptation through an automated modication of the visuomotor map that transforms visual inputs into motor outputs. b Post-trial visual

feedback results in adaptation that involves a cognitive process which offsets the presented target before utilising an unaltered visuomotor map

or task instruction (correct or do not correct movements online).

We also elected to investigate a potentially more sensitive means of examining the rates of adaptation since it was conceivable that cycle-averaging could mask subtle inter-group differences. We therefore also calculated b parameters on the basis of ts derived from the individual trials. Analyses based on these ts suggested that those participants who adapted using post-trial visual feedback did so more slowly than those participants who received online visual feedback. There was, however, no difference in the rate of performance improvement between the concurrent visual groups who were or were not permitted to make online corrections when moving towards visual targets. This particular result is consistent with a recent paper report by Franklin et al. (2008) who investigated adaptation to various force elds in the presence or absence of online visual feedback. They concluded that adaptation could occur in the absence of online visual feedback, but at a slower rate (determined by tting curves to individual trial data for each participant) compared to when online visual feedback was available.

Our results suggest that despite cycle-averaging being a useful tool in allowing assessment of task performance across all target conditions rather than only considering the rst/last trial of a particular epoch, care must be taken in interpreting the results, especially with respect to rates of adaptation derived from such data. Importantly, despite the potential caveat in interpreting the rates of adaptation in the present study due to the differing results obtained using the two methods, our interpretation of the results with respect to the expression of aftereffects remains robust. That is, we propose that adaptation to a novel visuomotor environment occurs principally by way of an automatic process when online visual feedback is available, but is mediated by the use of an explicit strategy when only post-trial feedback is provided.

As we alluded to in the Introduction, the isometric task that we reported in a previous study (e.g. Hinder et al. 2008) by its very nature, denied participants movement-related proprioceptive feedback. Yet in isometric contractions there are other rich sources of sensory feedback.

Muscle spindle sensitivity is greatest at low muscle forces similar to those we studied (Burke et al. 1978), although saturation may occur at higher levels of force (Rymer and Dalmeida 1980). In addition, Golgi tendon organ activity increases monotonically over a wide range of muscle force (Houk et al. 1981). Importantly, however, in isometric contractions (quiescent) antagonist muscles generate little such proprioceptive feedback. This is in marked contrast to dynamic task utilised in the present report in which spindle afferent signals are typically generated in both the actively shortening muscle and the lengthening antagonist muscle. We were interested in determining if these differences in the nature of the proprioceptive information lead to a change in the way in which adaptation occurred under the different visual feedback conditions. We also refer the interested reader to the work of Bernier et al. (2007) who used tendon vibration as a means of perturbing proprioceptive feedback.

Despite the differences in the nature of the proprioceptive feedback that was available in the isometric and dynamic tasks, the patterns of adaptation to the imposed rotation were similar, i.e., the presence of movement-related proprioceptive information in the dynamic task failed to alter the nding that post-trial visual feedback is an insufcient basis to permit automatic adaptation to occur. Rather than the specic nature of the proprioceptive feedback affecting adaptation, we found that temporal separation between proprioceptive feedback associated with an action, and visual feedback of the result of that action (i.e., detection of the error), appears to prevent automaticity in visuomotor adaptation. Bernier et al. (2006) discuss the issue of temporal separation of proprioceptive and visual feedback with respect to ballistic reaching movements. They suggest that visual information during movement leads to a more accurate estimate of interseg-mental dynamics of the moving limb i.e., one can determine the outcome of the immediately preceding motor command more accurately than in a delayed-visual feedback scenario. Indeed, our data suggest that an estimation of intersegmental dynamics thus derived promotes automatic adaptation of the visuomotor map, while post-trial feedback leads to adaptation via cognitive mechanisms,

123

206 Exp Brain Res (2010) 201:191207

whereby an automatic remapping does not necessarily occur.

McNay and Willingham (1998) considered the role of recalibration of the perceptuomotor map and strategies in visuomotor adaptation. Participants adapted to a 90 clockwise rotation of the (online) visual feedback during a line-tracing task. The extent of the adaptation attributable to the two proposed mechanisms was investigated in test trials in which participants were told that the rotation had been removed. Only performance improvements that occurred as a result of recalibrations might be expected to be apparent in that test, because an overt strategy could simply be abandoned if it was recognized as being inappropriate in the non-rotated context. McNay and Willinghams results indicated that throughout training, a combination of recalibration and strategic approaches were used, indicating that the performance in the novel context never became fully automated. This nding is in contrast to the results for our CF groups: while these participants did not exhibit full adaptation to the rotation, their aftereffects were of comparable magnitude to the performance improvements observed during training suggesting that the adaptation was fully automatic. Moreover, our results suggest that for the group who received only PF in our study, adaptation may have been mediated principally by the use of a strategy. If any proportion of the performance improvements exhibited by this PF group were mediated by an automatic recalibration, corresponding aftereffects would have been anticipated. However, no such aftereffects were observed. The contrast between the results of the present study, and those reported by McNay and Willingham (1998) suggest that the balance between automatic adaptation and strategic intervention is labile and task specic. As such, further work aimed at promoting an understanding of the factors that inuence this interplay would be benecial.

In summary, we have shown that in this visuomotor adaptation task, involving a novel two degree of freedom dynamic adaptation task, detection of sensory error drives feedforward adaptation, and that online corrections (feedback mediated modications) on the basis of the detected error do not necessarily improve the rate or the extent of the adaptation. These ndings replicate and extend those of Tseng et al. (2007), showing that their nding can be generalised to other dynamic visuomotor adaptation paradigms. Secondly, we have shown that for visuomotor adaptation to occur in an automatic manner (Cunningham 1989; Krakauer et al. 1999), detection of visual errors must occur concurrently with task execution. This nding expands on our previous work using an isometric task (Hinder et al. 2008), and suggests that regardless of the nature of the proprioceptive feedback that is available, it is

the concurrent nature of visual feedback that permits automatic visuomotor adaptation.

References

Bernier PM, Chua R, Franks IM, Khan MA (2006) Determinants of ofine processing of visual information for the control of reaching movements. J Mot Behav 38:331338

Bernier PM, Chua R, Inglis JT, Franks IM (2007) Sensorimotor adaptation in response to proprioceptive bias. Exp Brain Res 177:147156

Binsted G, Brownell K, Vorontsova Z, Heath M, Saucier D (2007)

Visuomotor system uses target features unavailable to conscious awareness. Proc Natl Acad Sci USA 104:1266912672Burke D, Hagbarth KE, Lofstedt L (1978) Muscle-spindle responses in man to changes in load during accurate position maintenance. J Physiol Lond 276:159164Caithness G, Osu R, Bays P, Chase H, Klassen J, Kawato M, Wolpert

DM, Flanagan JR (2004) Failure to consolidate the consolidation theory of learning for sensorimotor adaptation tasks. J Neurosci 24:86628671Cohen J (1969) Statistical power analysis for the behavioral sciences.

Academic Press, New YorkCunningham HA (1989) Aiming error under transformed spatial mappings suggests a structure for visual motor maps. J Exp Psychol Human Percept Perform 15:493506Ernst MO, Banks MS (2002) Humans integrate visual and haptic information in a statistically optimal fashion. Nature 415:429433 Flanagan JR, Rao AK (1995) Trajectory adaptation to a nonlinear visuomotor transformationevidence of motion planning in visually perceived space. J Neurophysiol 74:21742178 Franklin DW, So U, Burdett E, Kawato M (2008) Visual feedback is not necessary for the learning of novel dynamics. PLoS One 12:e1336Ghahramani Z, Wolpert DM (1997) Modular decomposition in visuomotor learning. Nature 386:392395Ghez C, Gordon J, Ghilardi MF (1995) Impairments of reaching movements in patients without proprioception. 2. Effects of visual information on accuracy. J Neurophysiol 73:361372 Hinder MR, Walk L, Riek S, Carson RG (2007) The interference effects of non-rotated versus counter-rotated trials in visuomotor adaptation. Exp Brain Res 180:629640Hinder MR, Tresilian JR, Riek S, Carson RG (2008) The contribution of visual feedback to visuomotor adaptation: How much and when? Brain Res 1197:123134Houk JC, Rymer WZ, Crago PE (1981) Dependence of dynamic-response of spindle receptors on muscle length and velocity. J Neurophysiol 46:143166Kawato M (1999) Internal models for motor control and trajectory planning. Curr Opin Neurobiol 9:718727Keppel G (1982) Design and analysis: a researchers handbook, 2nd edn, Prentice Hill, New JersyKlapp ST (1995) Motor response programming during simple and choice-reaction timethe role of practice. J Exp Psychol Human Percept Perform 21:10151027Klassen J, Tong C, Flanagan J (2005) Learning and recall of incremental kinematic and dynamic sensorimotor transformations. Exp Brain Res 164:250259Krakauer JW, Ghilardi MF, Ghez C (1999) Independent learning of internal models for kinematic and dynamic control of reaching. Nat Neurosci 2:10261031Krakauer JW, Pine ZM, Ghilardi M-F, Ghez C (2000) Learning of visuomotor transformations for vectorial planning of reaching trajectories. J Neurosci 20(23):89168924

123

Exp Brain Res (2010) 201:191207 207

Krakauer JW, Ghez C, Ghilardi M-F (2005) Adaptation to visuomotor transformations: consolidation, interference, and forgetting. J Neurosci 25:473478

Lackner JR (1974) Adaptation to displaced visionrole of proprioception. Percept Mot Skills 38:12511256

Lackner JR, Dizio P (1994) Rapid adaptation to coriolis-force perturbations of arm trajectory. J Neurophysiol 72:299313 Mazzoni P, Krakauer JW (2006) An implicit plan overrides an explicit strategy during visuomotor adaptation. J Neurosci 26:36423645

McNay EC, Willingham DB (1998) Decit in learning of a motor skill requiring strategy, but not of perceptuomotor recalibration, with aging. Learn Mem 4:411420

Miall RC, Jenkinson N, Kulkarni K (2004) Adaptation to rotated visual feedback: a re-examination of motor interference. Exp Brain Res 154:201210

Newell A, Rosenbloom PS (1981) Mechanisms of skill acquisition and the law of practice. In: Anderson JR (ed) Cognitive skills and their acquisition. Lawrence Erlbaum, Hillsdale, New Jersey

Pisella L, Grea H, Tilikete C, Vighetto A, Desmurget M, Rode G,

Boisson D, Rossetti Y (2000) An automatic pilot for the hand in human posterior parietal cortex: toward reinterpreting optic ataxia. Nat Neurosci 3:729736Rymer WZ, Dalmeida A (1980) Joint position sensethe effects of muscle-contraction. Brain 103:122Scheidt RA, Conditt MA, Secco EL, Mussa-Ivaldi FA (2005) Interaction of visual and proprioceptive feedback during adaptation of human reaching movements. J Neurophysiol 93:32003213 Shadmehr R, Mussa-Ivaldi FA (1994) Adaptive representation of dynamics during learning of a motor task. J Neurosci 14:3208 3224

Shemmell J, Tresilian JR, Riek S, Barry BK, Carson RG (2005)

Neuromuscular adaptation during skill acquisition on a two degree-of-freedom target-acquisition task: dynamic movement. J Neurophysiol 94:30583068Teasdale N, Bard C, Fleury M, Young DE, Proteau L (1993)

Determining movement onsets from temporal series. J Mot Behav 25:97106Tong C, Wolpert DM, Flanagan JR (2002) Kinematics and dynamics are not represented independently in motor working memory: evidence from an interference study. J Neurosci 22:11081113 Tseng YW, Diedrichsen J, Krakauer JW, Shadmehr R, Bastian AJ

(2007) Sensory prediction errors drive cerebellum-dependent adaptation of reaching. J Neurophysiol 98:5462van Beers RJ, Sittig AC, van der Gon JJD (1999) Integration of proprioceptive and visual position-information: an experimentally supported model. J Neurophysiol 81:13551364van Beers RJ, Wolpert DM, Haggard P (2002) When feeling is more important than seeing in sensorimotor adaptation. Curr Biol 12:834837Wang J, Sainburg RL (2004) Limitations in interlimb transfer of visuomotor rotations. Exp Brain Res 155:18Wigmore V, Tong C, Flanagan JR (2002) Visuomotor rotations of varying size and direction compete for a single internal model in motor working memory. J Exp Psychol Human Percept Perform 28:447457Wolpert DM, Ghahramani Z, Jordan MI (1995) Are arm trajectories planned in kinematic or dynamic coordinatesan adaptation study. Exp Brain Res 103:460470

123