Adaptation to gradual as compared with sudden visuo-motor distortions

Transcription

1 Exp Brain Res (1997) 115: Springer-Verlag 1997 RESEARCH NOTE selor&:florian A. Kagerer José L. Contreras-Vidal George E. Stelmach Adaptation to gradual as compared with sudden visuo-motor distortions csim&:received: 3 December 1996 / Accepted: 14 February p&:Abstract If visual feedback is discordant with movement direction, the visuo-motor mapping is disrupted, but can be updated with practice. In this experiment subjects practiced discrete arm movements under conditions of visual feedback rotation. One group was exposed to 10 -step increments of visual feedback rotation up to a total of 90, a second group to a 90 visual feedback rotation throughout the experiment. After the first group reached the 90 visual feedback rotation, its subjects performed faster, with less spatial error, and showed larger aftereffects than the subjects who practiced constantly under the 90 visual feedback rotation condition. Results suggest that gradually increasing feedback distortion allows more complete adaptation than a large, sudden distortion onset. dwk&:key words Visuomotor learning Internal models Visual feedback rotation Motor control&bdy: Introduction F. A. Kagerer J. L. Contreras-Vidal ( ) G. E. Stelmach Motor Control Laboratory, Arizona State University, Tempe, AZ , USA; Fax: +1 (602) , pepe@iris196.la.asu.edu&/fn-block: In visually guided limb movements a spatially perceived target direction is mapped into motor commands that drive the limb in the intended direction. This mapping between visual space and motor space is inappropriate if visual space is artificially rotated. In these circumstances movements become initially disordered with erratic movement paths. For example, a 90 rotation of visual space requires a movement direction that is perpendicular to the movement direction necessary when visual space and movement space are compatible. This results in a deviation of the movement path from a straight line. The movement initially starts in the direction of the visually perceived target, but as the lack of correspondence between visual feedback and performed movement becomes evident and the subject begins to update the visuo-motor mapping, corrections in the movement are made (Cunningham 1989). With practice, the initially erratic movement paths evolve as spirals and semicircles, and eventually appear as straight lines to the targets (Cunningham and Vardi 1990; Roby-Brami and Burnod 1995). The estimation of a rotational bias during a previous trial appears to induce an update of a correction factor at the sensory-motor map that is retained at the onset of the next trial. This adaptive process is based on the updating of the mapping between internal visual and motor representations of space which takes place over time, a process which is not easily overridden by conscious effort (Cunningham and Welch 1994). Previous studies involving visual-motor adaptation have been primarily concerned with the spatial properties of aiming movements during adaptation to changed execution conditions (Cunningham 1989; Bock 1992; Bock et al. 1992) rather than examining adaptation as a function of the time course of the perturbation. If the relative difficulty, e.g., measured by spatial error or movement time, of different transformations is preserved, i.e., if error patterns are invariant for adaptation to different degrees of perturbation, then the transformation-related error patterns may be attributed to structural (time-invariant) rather than dynamical (time-varying) properties of the system (Shadmer and Mussa-Ivaldi 1994). Our question is how a gradual increase in visual feedback rotation from 0 to 90 affects the adaptation of the visuo-motor map (a type of internal model), compared with a sudden rotation of 90. Recent experiments suggest that the central nervous system (CNS) uses sensory feedback to update an internal model of the interaction of the limb and the environment (Roby-Brami and Burnod 1995). As the CNS adapts to a new condition, the emphasis of the control strategy shifts from feedback toward feedforward control; as a result there is less need for on-line corrections. For example, in rotated visual feedback tasks, movement performance improvement correlates with a decrease of the initial directional error appearing immediately after movement onset, when visual feedback is not yet available. We hypothesized that gradual changes

2 558 in visual feedback rotation allow the system to estimate small rotational movement corrections that require only small updates in the visuo-motor map based on previously learned information (e.g., acquired before the rotation of visual feedback), while sudden large changes in feedback rotation initially require a search in visuo-motor space to minimize the error of performance with respect to the desired internal model parameters. According to this hypothesis it is predicted that a gradually increasing distortion of visual feedback would allow a better updating of the internal model than a sudden 90 distortion. We also expected larger aftereffects (movement paths mirroring the deviations found during practice) for the former, when the visual feedback rotation was suddenly removed after adaptation. Materials and methods Subjects Ten right-handed, young college students, who were randomly assigned to two groups of five subjects were tested. All subjects signed informed consent statements. Apparatus For data collection a Calcomp 9100 Series digitizing tablet (sampling rate 100 Hz) was used. Stimuli were presented on a computer screen [21 inch (53.5 cm), SVGA with points]. The targets were randomly presented in one of four different locations, around the home position in the center (see Fig. 2). Home position and target were visible over the whole duration of the trial. Experimental design Subjects made horizontal arm movements with a pen in their right hand on the digitizing tablet to the four different target directions that appeared one at a time on a computer screen in front of them (Fig. 2A); they had no vision of their moving arm and hand. Under a condition where visual feedback was rotated 90 counterclockwise, a movement away from the body towards the left on the horizontal tablet produced a downward left movement of the cursor on the vertical screen; analogous to that, a 10 rotation required a smaller change of movement direction relative to the cursor movement displayed.&fn.1:1 Subjects were instructed to make straight paths to the targets as fast as possible. Immediate feedback about movement time was given after each trial (Kagerer et al. 1995). To start each trial subjects had to click the pen briefly in the home position, wait for an acoustic stimulus appearing after a random delay of between 200 and 2000 ms, and start the movement from a home position toward the target upon appearance of the acoustic stimulus. An experimental session consisted of 11 blocks: one initial block of 40 movements aimed to four targets located at 45, 135, 225 and 335 (10 movements per target direction) as baseline condition (normal visual feedback), and 10 blocks of 60 move- 1 It should be noted that this arrangement required a second transformation besides the visual feedback rotation, namely a transformation from the horizontal plane of the movements to the vertical plane of the visual feedback. This second transformation was considered to have negligible influence on the subjects performance since all of them were used to working with a computer mouse, where the same transformation has to be made. Moreover, the same setup was used for both groups.&/fn: ments (15 per target direction) as practice condition (rotated visual feedback). Following adaptation, 16 movements (four per target direction) were administered under baseline conditions to test for aftereffects; two subjects in each group performed only four movements (one per target direction). The order of appearance of the four target locations was randomized within each block, but each target appeared the same number of times. One group was presented with a 90 visual feedback rotation from the first block of practice onward (S: sudden onset group), whereas a second group was presented with a blockwise, monotonic increase in rotation of visual feedback by 10 increments, until the rotation had reached 90 (G: gradual onset group). This block, containing the largest and most difficult visual feedback distortion, was administered twice. Both groups performed the same number of trials. Data acquisition Movement trajectories were recorded by sampling the position of a pen subjects held in their right hand; a cursor on the screen moving with unitary gain relative to the pen movements provided the visual feedback. Recording began with the acoustic start signal and ended with the second acoustic signal upon arriving at the target. Movement time (MT) was calculated as the difference in time between leaving the home position and entering the target. Straightness of the movements was measured by calculating the root mean squared error (RMSE), defined as RMSE = where (x a, y a ) are points of the actual trajectory, (x i, y i ) the corresponding points on an ideal, straight line between home position and target, and N is the number of samples (Cunningham 1989). The initial directional error (IDE) was measured as the angular directional difference between movement path and an ideal straight line between home position and target; IDE was assessed during the first 80 ms of movement to make sure that no visual feedback signals could be utilized. Data analysis The collected data were filtered off-line with a 10 Hz lowpass filter, and RMSE, MT and IDE were calculated. For analysis, the movements toward the four target locations belonging to the same trial were pooled and one mean was calculated, resulting in 164 data points per subject (Fig. 1A). A Group Trials repeated-measures ANOVA was performed for RMSE, MT and IDE for the last two practice blocks. Since only four observations were taken into account, aftereffects were analyzed using the t-test. Results L NM 2 2 ( xa xi) + ya yi N a During the baseline condition there was no statistically significant difference between the groups. Repeatedmeasures ANOVA (Groups Trials) showed no significant group differences for RMSE, MT and IDE (P>0.05). As expected, in the practice condition the two groups showed rather different performance curves. The sudden onset condition resulted in a severe disruption of the straightness of movements, as shown by the larger RMSE at the beginning of practice (Fig. 1A, see also Fig. 2A C for the movement paths), decreasing exponentially with practice and approaching a baseline level; MT and IDE also decreased in a similar way with practice. f O QP 05.

3 559 Fig. 1A D Performance curves of the two groups. A The minimal disruption of performance, as measured by spatial error, under conditions of gradually increasing distortion of visual space (group G: thin line), and the consistently larger root mean square error (RMSE) resulting from the sudden onset of the distortion (group S: bold line). The characteristics of this performance curve were also present in the curves for initial directional error (IDE) and movement time (MT) (not shown). The first ten trials show the baseline performance. Each of the longer ticks on the x-axis marks the beginning of a block, which meant an increase by 10 for group G, while group S was always presented with the same visual feedback rotation of 90. The empty arrows at the right side of the graph mark the aftereffect, which was less pronounced in group S than in group G. B The last 120 practice movements (pooled to the 30 trials shown), where both groups were exposed to the 90 feedback rotation. C,D The groups performance for IDE and MT for the last two blocks of practice&/f :c.gi The gradual onset condition caused little disruption of the subjects performance over the whole range of distortions (Figs. 1A, 2B,C). Linear regression showed a significant correlation between increasing RMSE and trial number, indicating the increasing difficulty of the transformation (r=0.66, slope=0.02, P<0.01). Within blocks, in group G only the first two trials of each block showed a slight increase in RMSE, the values thereafter dropping immediately to baseline level. Although the performance curves for group S approach those for group G, the two groups show clear differences in performance at the 90 rotation condition (last two shaded blocks in Fig. 1B). Subjects in group G performed better than subjects in group S (who at that point had practiced the 90 rotation for 480 trials) in terms of RMSE, MT and IDE. A repeated-measures ANOVA (Groups Trials) showed that in the last block, when the visual feedback rotation was identical for the two groups, group G showed a significantly smaller RMSE (F 1,8 =7.03, P<0.05) than group S (Fig. 1B). A linear fit showed a significant negative correlation between RMSE and trial number over the last two blocks (r= 0.56, slope= 0.08, P<0.001) for group G, which suggests a further reduction of RMSE during the final trials. Group S, on the other hand, showed a higher RMSE, and a linear fit revealed no tendency for further improvement (r=0.03, slope= 0.005, P>0.05), suggesting that the subjects did not update their internal model during the last 120 movements. Another indicator of the adaptation is the IDE, measured 80 ms after movement onset. If a movement starts in the deviant direction, its RMSE is likely to be larger, assuming that subjects use the strategy of moving in a curved path rather than using a straight path correction. Repeated-measures ANOVA showed that in the last two blocks there was a significant main effect for group (F 1, 8 =7.11, P<0.05), indicating that the IDE of group G was significantly smaller than the IDE of group S. A linear fit showed a significant negative correlation between IDE and trials (r= 0.6, slope= 0.2, P<0.001) and no correlation for group S (r=0.03, slope=0.06, P>0.05). As Fig. 1C illustrates, in the last block IDE of group G is closer to the desired value of zero than that of group S (mean values of 6.4 vs 13.45, respectively). Our results showed that IDE correlated very well with RMSE (group G: r=0.83, P<0.001; group S: r=0.93, P<0.001). We found a similar trend for MT during the last two blocks. Since aftereffects are usually very transient and dissipate quickly, we compared RMSE and IDE only for the first four post-practice movements (baseline condition) for each group. The movement paths of one subject for each group, depicted in Fig. 2D, represent the group results very well. Both groups showed an aftereffect and moved initially in the direction of the rotated feedback. However, t-tests revealed that RMSE and IDE for group G during post-practice performance under normal visual feedback were significantly higher than those for group

4 :b.lbt/& 560 Table 1 Aftereffect: Mean root mean square error (RMSE), initial directional error (IDE) and standard deviations for the first four trials under normal visual feedback condition after adaptation. With one exception, the subjects of the sudden onset group (S) show a smaller RMSE than those of the gradual onset group (G). Subjects 2 and 3 of each group performed only one trial, therefore there is no standard deviation. Positive IDE values denote clockwise, negative values counterclockwise deviation from a straight line to the target&/tbl.c:&tbl.b: Gradual onset group (G) Sudden onset group (S) Subjects RMSE [mm] IDE (deg) RMSE [mm] IDE (deg) (2.8) 55.3 (4.6) 20.4 (0.9) 49.2 (3.2) (8.4) 30.6 (14.8) 18.7 (9.4) 33.4 (19.1) (1.6) 43.0 (5.9) 6.1 (2.3) 1.3 (5.4) Fig. 2A D Movement paths of one subject of each group; the first four trials of a block are shown. A Home position (HP) and targets (T) are displayed. Subjects started from the home position, moving to the target (1 1 cm) displayed on the monitor. Left-hand column: A baseline performance, B performance under 10 visual feedback rotation and C performance under 90 visual feedback rotation (last but one block, which was the first block with 90 rotation for group G). Variability is small, and the curvature under the 90 condition is just slightly higher than during baseline performance. Right-hand column: A baseline performance, B performance under 90 visual feedback rotation (first practice block) and C 90 visual feedback rotation (last but one block). D Aftereffects are more pronounced after gradual than after sudden onset practice, suggesting a more elaborated visuo-motor map for group G&/f :c.gi S (RMSE: t=7.43, P<0.001; IDE: t=6.15, P<0.001) (Table 1). Discussion The results suggest that the smaller RMSE and IDE, as well as the shorter MT, for group G may result from a more elaborated updating of the visuo-motor map due to exposure to a slowly increasing visual feedback rotation. Our results on the aftereffects are in agreement with this view. The aftereffect can be seen as an indicator of the degree to which the internal model has been updated during adaptation. The first few post-practice trials under normal visual feedback condition are performed under the previously adapted visuo-motor map. Therefore, a more elaborated visuo-motor map would cause a larger RMSE and IDE, and longer MT. Since at late stages of adaptation the movement path usually takes the form of a shallow arc (Fig. 2D, sudden condition), and since this kind of movement usually shows only a little variability in the tangential velocity, it can be assumed that a slightly curved line can be executed as fast as a straight line. Therefore, straightness of a path measured as RMSE with reference to baseline performance provides information about the degree to which the subject s internal model has adapted to the rotated feedback. IDE may be considered as an indicator for the visuo-motor transformations in the initial phase of the movement, as feedback is not likely to have an influence then. These results suggest that gradually increasing visual feedback rotation allows a more complete adaptation to this distortion, with minimal disruption in performance, compared with a sudden large distortion onset. Since the increase in the distortion was stepwise, one could argue that subjects anticipated a change with every new block. The increase in RMSE at the beginning of each block, however, does not support this argument. Also, at the end of the experiment all subjects in group G reported that they did not notice any systematic change during practice, except during the aftereffect trials. These findings may represent a generalized feature of adaptation mechanisms that may be applicable to other conditions which require adaptation to new environments. We have presented evidence that the application of a distortion that evolves gradually to its desired final level results in minimal degradation of performance compared with a single large distortion, and in a more elaborated sensorimotor mapping, as shown by the larger aftereffects. 2.p&:Acknowledgements This research was supported by a Flinn Foundation Grant in the Computational Aspects of Motor Adaptation. We thank our colleagues at the Motor Control Laboratory at ASU for their comments on a earlier draft of this manuscript.

5 561 References Bock O (1992) Adaptation of aimed arm movements to sensorimotor discordance: evidence for direction-independent gain control. Behav Brain Res 51: Bock O, Howard IP, Money KE, Arnold KE (1992) Accuracy of aimed arm movements in changed gravity. Aviat Space Environ Med 63: Cunningham HA (1989) Aiming error under transformed spatial mappings suggests a structure for visual-motor maps. J Exp Psychol Hum Percept Perform 15: Cunningham HA, Vardi I (1990) A vector-sum process produces curved aiming paths under rotated visual-motor mappings. Biol Cybern 64: Cunningham HA, Welch RB (1994) Multiple concurrent visualmotor mappings: implications for models of adaptation. J Exp Psychol Hum Percept Perform 20: Kagerer F, Bloedel JR, Stelmach GE (1995) Learning of visuomotor coordination with rotated visual feedback. Soc Neurosci Abstr 21: 1443 Roby-Brami A, Burnod Y (1995) Learning a new visuomotor transformation: error correction and generalization. Cogn Brain Res 2: Shadmer R, Mussa-Ivaldi RA (1994) Adaptive representation of dynamics during learning a motor task. J Neurosci 14: