- Email: [email protected]
Adapting to target error without visual feedback
Acta Psychologica 143 (2013) 129–135
Contents lists available at SciVerse ScienceDirect
Acta Psychologica journal homepage: www.elsevier.com/ locate/actpsy
Adapting to target error without visual feedback Brendan D. Cameron a,⁎, Jarrod Blinch b, Alyson Plecash b, Jordan Squair b, Lauren Wou b, Romeo Chua b a b
Departament de Psicologia Basica, Universitat de Barcelona, Barcelona, Catalonia, Spain School of Kinesiology, University of British Columbia, Vancouver, Canada
a r t i c l e
i n f o
Article history: Received 16 August 2012 Received in revised form 28 January 2013 Accepted 4 March 2013 Available online 2 April 2013 PsycINFO classification: 2330 Keywords: Adaptation Reaching Saccade Online control
a b s t r a c t What information is necessary for the motor system to adapt its behaviour? Visual hand-to-target error provides salient information about reach performance, but can learning proceed without this information? We investigated adaptation to an unperceived target perturbation under visual open-loop conditions. Participants looked and reached, without any vision of their hand, to a target that jumped rightward at saccade onset (Perturbation condition) or remained stationary throughout the trial (Stationary condition). The target jump in the Perturbation condition was tied to the saccade, such that participants were unaware that it had occurred. Each type of exposure was followed by a posttest, in which participants reached to a target that disappeared at saccade onset. In the posttest, participants reached farther following exposure to the perturbation than they did following exposure to the stationary target, indicating that participants had learned from systematic exposure to the jump. These findings imply that online error induces motor learning, even when participants receive no visual information about their performance. © 2013 Elsevier B.V. All rights reserved.
1. Introduction One of the ways that we refine our movements is by comparing the endpoint of a movement to a desired endpoint. When we reach for and miss a target, for instance, a visual error signal can update commands for future reaches (Magescas & Prablanc, 2006). But what if there is no visual information about the success or failure of a reach? Can the motor system use an internal estimate of limb position to adaptively modify subsequent reaches? The present study addressed this question. We already know that visual feedback is not required for real-time motor responses to target error. If a target jumps during a reach, the unseen hand will automatically deviate toward it, even when the participant is unaware of the jump (Goodale, Pelisson, & Prablanc, 1986; Prablanc & Martin, 1992). The online correction that occurs under these conditions may be driven by a proprioceptively-derived estimate of limb position or it may be driven by a forward-model-derived estimate of limb position, wherein a copy of the motor command is used to predict the limb's future position. Arguments for the involvement of forward models in these online corrections are based on the rapidity with which appropriate corrections can occur and on the observation that a deafferented patient was able to perform such corrections (Bard et al., 1999; Desmurget & Grafton, 2000). However, it is not clear if the online error signal described above operates only in real-time, to drive the online correction, or whether ⁎ Corresponding author. Tel.: +34 93 312 5143. E-mail address: [email protected] (B.D. Cameron). 0001-6918/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.actpsy.2013.03.002
it also acts as a training signal, modifying subsequent motor performance. Recent work from our group showed that unseen online target jumps produced motor adaptation, even though the online corrections eliminated terminal error (Cameron, Franks, Inglis, & Chua, 2011). In that study, we exposed participants to repeated rightward target jumps during reaches with an unseen hand, and we observed adaptation to the perturbation (cf. Magescas, Urquizar, & Prablanc, 2009). However, while vision of the hand was not available during the reach in the study of Cameron et al. (2011), vision was reintroduced at the end of each reach. So, although online correction of the reach eliminated terminal error between the hand and the target, there was visual confirmation that the target had been acquired. The present study tested for adaptation when neither online nor terminal visual feedback was available. That is, we tested whether people developed reach aftereffects after aiming to a target that imperceptibly jumped during their movement, in the absence of any real-time or terminal visual feedback. Such stimulus conditions are reminiscent of those used to induce saccadic adaptation, where the gain of saccades is adaptively altered by systematically changing the location of the saccade target while the eyes are moving (e.g., Deubel, Wolf, & Hauske, 1986; McLaughlin, 1967; Wallman & Fuchs, 1998). When the initial saccade is complete, a different distance is present between the foveated location and the location of the target than was anticipated at the start of the saccade. After repeated exposure to such error, primary saccades gradually become larger or smaller, depending on the direction of the perturbation. A key difference between motion of the eyes and the hand, however, is that the former are capable of very little, if any, online control, whereas
130
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
the latter is capable of online corrections that can eliminate some or all of the initial error in the movement. Indeed, Magescas and Prablanc's (2006) original study demonstrating adaptation of the hand to target perturbations had a protocol designed precisely to prevent any online corrections to the reach; the target was presented at its perturbed location only after completion of the reach, mimicking the error signal that is present during saccadic adaptation. Magescas and Prablanc (2006) demonstrated robust adaptation of reaching movements under such conditions, showing that terminal visual error – in the absence of any visual perturbation of the effector – was sufficient to drive adaptation. In the present study we considered a contrasting set of conditions, where online correction was possible and no visual error signal was available, to investigate whether the error signal that drives real-time reach corrections might also produce adaptive changes in the motor system. We emphasize that the present study was designed to test the role of an online error signal in the training of subsequent movements. It did not allow us to determine whether the nature of the error signal was proprioceptive and/or efference copy-based, and we made no assumptions about which of these signals was potentially driving adaptation. 2. Materials and methods 2.1. Participants Eight participants from the university community (4 male, 4 female; ages 20–22) completed the study. All participants were self-described right-handed, had normal or corrected-to-normal vision, and were naïve to the aims of the study. All participants provided informed consent prior to the experiment, and the study was conducted in accordance with the guidelines of the university's research ethics board. 2.2. Apparatus Stimuli were projected onto a half-silvered mirror mounted midway between a horizontal reaching surface and an inverted LED array, such that targets (LEDs) appeared to be in the same plane as the reaching hand. Participants rested their head in a chin-rest, such that their eyes were 50 cm from the reaching surface. Vision of the hand was occluded when a white light below the mirror was extinguished. The apparatus was located in a dark room. Electrooculography (EOG) was used to monitor horizontal saccades. Disposable Ag–AgCl surface electrodes were placed at the outer canthi of the eyes with a reference electrode on the forehead. EOG signals were amplified (5–10 K) and band-pass filtered (0.1–30 Hz) with an AC preamplifier (Grass Instruments P511). The experimenter manually set a voltage threshold for each participant such that the trigger would occur within approximately the first half of the saccade. The first peak in the EOG signal was interpreted as completion of the primary saccade. The experimenter used an onscreen display of the EOG signal to visually estimate the magnitude of the participant's saccades and then set the voltage threshold accordingly. Precise calculation of saccade magnitude was done offline. Hand movements were recorded with Optotrak (Northern Digital), sampling at 500 Hz. An infrared emitting diode was mounted on the tip of a hand-held stylus. The stylus was also equipped with a pressure sensitive tip that was used to record movement onset and end. The experiment was run with a dedicated computer running DOS, with a digital I/O card with high-resolution timers (10 kHz). For control of the target jump within a trial, the analog EOG signal was fed through an independent hardware analog circuit that, when the EOG signal exceeded a set threshold, set a transistor–transistor logic trigger. This trigger (square wave) signal was used to trigger the LED onset/offset for the target jump. The delay between this trigger being set and the LED being triggered was less than 1 ms.
2.3. Task Participants began each trial with the stylus placed at the home position and their eyes on the fixation point, 18.5 cm above the home position. After a variable foreperiod a target would appear 21 cm to the right of fixation (Center location), coincident with the offset of fixation. On look-and-reach trials (indicated by a red fixation point), the participant's task was to look and reach to the target in a single, smooth, and accurate motion. If movement time was faster than 300 ms or slower than 600 ms, participants were asked to slow down or speed up on the next trial, accordingly. On look-only trials (indicated by a green fixation point), the participant's task was to only look at the target, keeping their hand at the home position. Look-only trials were interleaved with reaching trials in the Practice and Exposure phases of the experiment, described next. 2.4. Conditions and phases Each participant completed two conditions on the same day, an experimental condition (Perturbation) and a control condition (Stationary), the order of which was counterbalanced across participants. Each condition consisted of 4 phases: Practice (12 look-and-reach trials, 12 look-only trials), Exposure (35 look-and-reach, 35 look-only), Posttest 1 (20 look-and-reach), and Posttest 2 (10 look-and-reach), in that order. In the Practice phase, participants received full vision of their hand and the target throughout each trial. In all other phases, vision of the hand was available at the start of the trial, but then extinguished for the entirety of the reach (lights off at saccade start). One second after completion of the reach, the target was extinguished (if it was present) and a tone was sounded, indicating to participants that they could return their hand to the home position. Then, 750 ms after the participant began to return to the home position, vision of the hand was re-introduced to allow precise placement of the stylus at the home position. The only difference between the Perturbation and Stationary conditions occurred in the Exposure phase. In the Perturbation condition, the target jumped during the Exposure phase: 4.7 cm right (from Center to Right location) on look-and-reach trials and 4.7 cm left (from Center to Left location) on look-only trials. We included lookonly trials in order to inhibit the accumulation of any saccadic adaptation on look-and-reach trials (Cameron et al., 2011; Magescas et al., 2009). If such adaptation occurred, it might transfer to the reaching limb (Bekkering, Abrams, & Pratt, 1995). The target jump, which consisted of extinguishing an LED at one location and immediately illuminating an LED at another location, was triggered during the saccade to eliminate any awareness of the jump (Bridgeman, Hendry, & Stark, 1975). In the Stationary condition, the target did not jump during the Exposure phase. Instead, it initially appeared at the Right location and remained at that location throughout the reach (look-and-reach trials) or initially appeared at the Left location and remained at that location throughout the eye-movement (look-only trials). On Posttest 1 trials the target appeared at the Center location, then disappeared at saccade start and did not reappear. This first posttest, in which no target was present during the reach, was designed to detect any changes in movement planning that resulted from exposure to the perturbation. If adaptation was to accumulate, it should manifest in Posttest 1 as positive endpoint bias in the Perturbation condition relative to the Stationary condition. This contrasts with Posttest 2 trials, in which the target appeared at the Center location and remained lit until 1 s after movement completion. This second posttest, in which a stationary target was visible during the reach, was designed to wash out any accumulated adaptation. Here, we would expect online correction to the stationary target and an unlearning of any adaptation that may have occurred during exposure. Our protocol was designed for a comparison between the Perturbation and Stationary conditions in the posttest phases, rather than a
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
comparison between a pretest and posttest phase in the Perturbation group. We deliberately excluded pretests from our design. In prior experiments where we extinguished hand and target vision during reaching we observed sizeable drifts in movement endpoint over the course of reaching: participants increasingly undershot the target as trials progressed (Cameron, Franks, Inglis, & Chua, 2010; Cameron et al., 2011). Accordingly, had we included a pretest in the current study, we would have expected drift to begin in that phase and continue to accumulate throughout subsequent phases. With recalibration not possible during the no-vision exposure condition, we would expect drift-driven undershooting to be larger in the posttest than in the pretest, thus masking some or all of the adaptation to the target error. In the current design the Stationary and Perturbation conditions should accumulate equivalent amounts of endpoint drift, and comparison between them should consequently reveal any target-driven adaptation in the Perturbation condition.
131
reach onset – a measure intended to index the prepared response – no significant effects emerged (p > .10). It is not clear whether an absence of a significant perturbation effect at this early time point suggests that the aftereffect did not extend to motor preparation (and, rather, was restricted to online control), or whether the effect of the perturbation on motor preparation was simply too small to detect at such an early stage of the reach. Because the visual target was extinguished at reach onset in Posttest 1, we favour the view that performance in that phase reflected a prepared response. The velocity profiles in the early posttest (Section 3.3), which suggest a more robust response in the perturbation condition, also support this interpretation; however, we cannot rule out the possibility that some or all of the effects we have observed result from an influence of the perturbation on online control only. 3.2. Real-time response to the perturbation: performance in the Exposure phase
2.5. Analysis To assess the effects of the perturbation we examined radial position of the hand relative to the home position. We examined the position of the hand at 3 time points: reach endpoint, 200 ms after reach onset, and 100 ms after reach onset, the last of which was intended as an index of the prepared movement. For each participant, position averages were calculated for every “block” of 5 consecutive trials. For the Exposure phase we examined the hand's position at each time point (reach endpoint, 200 ms, and 100 ms) with separate 2 Condition × 7 Block repeated measures ANOVAs. For Posttest 1 we applied 2 Condition × 4 Block repeated measures ANOVAs for each time point. The Huynh–Feldt correction to degrees of freedom was applied in the case of sphericity violations, which were detected with Mauchley's at p b .10. Statistical analyses were carried out with the ‘ez’ package (Lawrence, 2012) in R (R Core Team, 2012). We obtained the magnitude of the primary saccade on each trial by first locating, in the EOG trace, the first positive voltage peak after the onset of the target. We worked backwards from this peak to locate the first local minimum (the onset of the saccade). The voltage difference between these two points was used as a measure of saccade magnitude. For statistical analysis, participants' mean voltage values were submitted to a 2 Condition × 7 Block repeated measures ANOVA in the Exposure phase and a 2 Condition × 4 Block repeated measures ANOVA in Posttest 1. 3. Results and discussion 3.1. Adaptation to the perturbation: performance in the Posttest 1 phase To determine if people adapted to the perturbation, we first compared movement endpoints from the Posttest 1 phases of the Perturbation and Stationary conditions. If the motor system implicitly learned from exposure to the target jump, we should observe farther reach endpoints in the posttest of the Perturbation condition than in the posttest of the Stationary condition. As shown in Fig. 1 (top), participants reached farther after exposure to the perturbation than after exposure to a stationary target, F(1,7) = 32.78, p b .001. We also observed a condition × block interaction, F(3,21) = 3.35, p = .04, indicating that the endpoint difference between conditions was not constant across all blocks of the posttest. The size of the difference decreased after the first block of the posttest (Fig. 1, top). Earlier stages in the reach also showed a pattern of farther reaching in Posttest 1 for the perturbation condition than for the stationary condition (Fig. 1, middle and bottom panels). At 200 ms after reach onset there was a significant effect of condition, F(1,7) = 7.87, p b .05, and a marginally significant interaction, F(3,21) = 2.91, p = .058, suggesting, like the endpoint results, that the aftereffect of the perturbation was strongest early in the posttest. At 100 ms after
Analysis of endpoint performance in the Exposure phase suggested that participants were largely effective at responding online to the target jump in the Perturbation condition (Fig. 1). Endpoints did not differ significantly between the two conditions in the Exposure phase, F(1,7) b 1. Recall that the target jumped from the Center to the Right location in the Perturbation condition, whereas it was always present at the Right location in the Stationary condition. So, if participants had failed to respond online to the jump, a relatively large difference (~3.7 cm in radial distance) would have been present between the endpoints for the two conditions in the Exposure phase. Furthermore, when the 35 look-and-reach trials of the Exposure phase were considered as 7 consecutive blocks of 5 trials, there was no interaction between block and condition, F(6,42) b 1. However, an absence of a significant difference here does not necessarily mean that the endpoints were equivalent between conditions. Some undershooting may have been present in the perturbation condition that we were not able to detect statistically. Although the F-value for the condition comparison was very small (less than 1) for the radial distance measure, it was larger for the lateral position measure (condition effect: F(1,7) = 3.11, p = .12, data not shown). In other words, there may have been a trend in the Exposure phase for more leftward endpoint error in the Perturbation condition than in the Stationary condition, and we cannot rule this out as a possible contributor to the adaptation. Radial distance at 100 ms after reach onset is a measure that should be free from contamination by online processing. Accordingly, distance at 100 ms, tracked across blocks of trials, should tell us about changes in movement preparation in the Exposure phase caused by repeated perturbations (Fig. 1, bottom). We anticipated a constant distance across blocks for the Stationary condition (for which the target was at the same location from the start and end of each trial) and a gradually increasing distance across blocks for the Perturbation condition. Such an effect would manifest as an interaction between the condition and block factors. However, we observed only an effect of block, F(2.6,17.9) = 5.56, p b .01, indicating a decrease in distance travelled for both conditions over blocks. Neither the interaction effect, F(6,42) = 1.99, p = .09, nor the condition effect, F(1,7) b 1 was significant. The absence of the predicted interaction effect at 100 ms may indicate, as discussed in the previous section, that insufficient time had passed for the effect to emerge. At 200 ms, a significant interaction effect was present, F(6,42) = 2.53, p b .05, though it is unclear whether its pattern (Fig. 1, middle) is consistent with what we had predicted. There was also a significant effect for block, F(6,42) = 5.34, p b .001, but no effect for condition, F(1,7) b 1. To summarize: the endpoint analysis of the Exposure phase suggested that participants corrected their movements online to the perturbation, reaching endpoint locations that were comparable, if not necessarily equivalent to those reached when the target was
132
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
320 280
300
Stationary Perturbation
260
Radial distance from start (mm)
Reach endpoint (mean MT = 394 ms)
Exposure
Post1
Post2
180
200
220
Stationary Perturbation
160
Radial distance from start (mm)
200 ms after reach onset
Exposure
Post1
Post2
80 50
60
70
Stationary Perturbation
40
Radial distance from start (mm)
100 ms after reach onset
Exposure
Post1
Post2
Fig. 1. Radial distance (relative to the home position) travelled by the stylus at reach endpoint (top panel), 200 ms after reach onset (middle panel), and 100 ms after reach onset (bottom panel). Each data point represents the mean of 5 consecutive trials. In the top panel, the upper horizontal dashed line indicates the radial distance of the Right target location, while the lower horizontal dashed line indicates the radial distance of the Center target location. Error bars denote 1 standard error of the mean. Note the different scaling of the y-axis for the bottom panel.
stationary at the farther location. The analysis of distance travelled at 100 ms and 200 ms provided ambiguous results. There was some indication from the interaction effect for the 200 ms measure that distance travelled increased for the Perturbation condition relative to the Stationary condition, but the pattern was not clear, and we accordingly refrain from claiming that an adaptation effect was observable within the Exposure phase (though we do think the presence of adaptation can be clearly inferred from the results of the posttest described in the previous section). 3.3. Velocity profiles in the Exposure and Posttest 1 phases We examined the velocity profiles at the start (first 5 trials) and end (last 5 trials) of the Exposure and Posttest 1 phases (Fig. 2). These profiles suggest comparable performance for Stationary and Perturbation conditions in the Exposure phase, but larger amplitude movements for the Perturbation condition than the Stationary condition in the early Posttest 1 phase. Analysis of the peak velocities at each of these stages revealed a significantly larger peak velocity for the Perturbation condition in early Posttest 1, t(7) = 3.97, p b .01, but no significant differences between the conditions in early Exposure, t(7) = −.78,
p > .20, late Exposure, t(7) = 1.35, p > .20, or late Posttest 1, t(7) = 1.66, p > .10.
3.4. Saccadic adaptation was not responsible for reach adaptation Primary saccade amplitudes in Posttest 1 did not differ significantly between the Stationary condition (mean: 2.5 V) and the Perturbation condition (mean: 2.3 V), F(1,7) = 1.37, p = .28. This implies that the opposite direction target jumps on the saccade-only trials during the Exposure phase successfully inhibited any accumulation of rightward saccadic adaptation, consistent with previous studies that have used this method (Cameron et al., 2011; Magescas et al., 2009). Accordingly, any difference in reach performance between the conditions is not likely to be due to saccade-related processes. That is, farther reaching in the Perturbation condition was not a consequence of saccade magnitude. In the Exposure phase, by contrast, primary saccade amplitude was larger in the Stationary condition than in the Perturbation condition, F(1,7) = 32.06, p b .001. This was expected, as the initial target location was 4.7 cm farther right in the look-and-reach trials of the Stationary Exposure phase, and it serves as confirmation that our EOG
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
Exposure End 1000 600
Velocity (mm/s)
600 200
Stationary Perturbation
0
200
1000
Stationary Perturbation
0
Velocity (mm/s)
Exposure Start
0
100
200
300
400
0
Time after reach onset (ms)
100
200
300
400
Time after reach onset (ms)
Post1 End
600
Velocity (mm/s)
600 200
0
0
Stationary Perturbation
200
1000
Stationary Perturbation
1000
Post1 Start
Velocity (mm/s)
133
0
100
200
300
400
Time after reach onset (ms)
0
100
200
300
400
Time after reach onset (ms)
Fig. 2. Mean resultant velocity profiles at 4 time points: the first 5 trials (Start) and last 5 trials (End) of the Exposure and Posttest 1 phases. To obtain the displayed means, trajectories were locked to movement onset, cropped at 400 ms, and collapsed across all participants.
measure of saccade magnitude was sensitive to target location, as we have shown previously (Cameron et al., 2011). 3.5. Participants were unaware of the target jump Each participant was asked at the end of the experiment whether they saw the target jump. Two participants thought they might have seen the target jump one time during the experiment. (It jumped 70 times). All other participants reported no awareness of the jump. Furthermore, when asked whether they noticed a difference between the two sessions of the experiment (i.e., Stationary vs. Perturbation), all participants reported no difference. This lack of awareness of the jump is consistent with previous studies employing comparable saccadically-triggered target jumps (e.g., Bridgeman et al., 1975; Cameron et al., 2011; Goodale et al., 1986; Magescas et al., 2009; Prablanc & Martin, 1992). 3.6. Timing of the target jump with respect to the eye and hand Here we report the relative timing of the eye, hand, and perturbed target onsets in the Exposure phase of the Perturbation condition. (Group means are presented, followed by mean within-subject standard deviations in brackets.) When participants looked and reached for the target, the eye preceded the hand by 93.6 ms (73.7), and the target jump was triggered 33 ms (6.3) after the onset of the saccade. Mean saccade duration in this condition was 76.3 ms (7.7), and mean movement time of the reach was 385.8 ms (18.7). 4. General discussion In our experiment, people adapted to an unperceived target jump without any visual feedback. This suggests that error that arises during a reach – error that the motor system compensates for in real-time – informs performance on subsequent reaches. Moreover, it suggests
that learning occurs even when this error signal relies on non-visual estimates of effector position. Motor adaptation studies have traditionally manipulated either the force acting on the limb, via a robotic arm (e.g., Shadmehr & MussaIvaldi, 1994) or Coriolis forces (e.g., Lackner & Dizio, 1994), or the visual feedback of the limb, via prisms (e.g., Welch, 1986) or cursor rotations (e.g., Krakauer, Pine, Ghilardi, & Ghez, 2000). The present study contrasts with such studies, as neither the force acting on the limb nor vision of the limb was altered. This is in line with work by Magescas and Prablanc (2006), who demonstrated adaptation to terminal reach error induced by a target perturbation. In that study, however, the critical adaptive signal was visual hand-to-target error. Our results suggest that reach error needs to be neither terminal nor visual in order for adaptation to target error to occur. One possibility that we have not yet addressed is that the online correction itself, rather than the error signal that drives it, is responsible for the adaptation that we observed. While the present study cannot distinguish between these two possibilities, previous studies of reaches (Tseng, Diedrichsen, Krakauer, Shadmehr, & Bastian, 2007) and of saccades (Wallman & Fuchs, 1998) suggest that it is the error signal, rather than the corrective motor response, that is primarily responsible for adaptation. We, accordingly, favour the argument that the real-time error signal, rather than the motor correction, was responsible for the adaptation observed in the present study. However, the studies of Tseng et al. and Wallman and Fuchs tested a visual error signal (absent in the present study), and so the applicability of their findings to the present study is not clear. A future study applying the present study's visual manipulations while testing both rapid, uncorrectable movements and slower, correctable ones (a manipulation employed by Tseng et al., 2007) might address this question. Though we have suggested that our results indicate an effect of online error (and/or online correction) on reach adaptation, an alternative explanation is that terminal feedback provided the critical signal for adaptation. Although participants in our study did not receive any visual feedback at the end of their reach, proprioceptive feedback was always
134
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
available, and it may have permitted a comparison between the visual target and the hand's location at the end of the reach. Movement endpoints did not differ significantly between the Perturbation and Stationary conditions during the Exposure phase, and we took this as evidence that terminal error signals were comparable between the two conditions, arguing in favour of the online error signal in the Perturbation condition being the critical difference. However, our argument works only if one considers a spatial error signal to be the sole driver of motor adaptation. A terminal reward signal, rather than a spatial error signal, could be responsible for the effects we observed. Izawa and Shadmehr (2011) recently showed that a reward signal, in the absence of any visual feedback, is able to produce adaptation of reaching movements. In the reward condition of that study, participants were only provided with an indication of success or failure of the reach (the target exploded if the reach was successful). They compared this to conditions where visual feedback of a rotated cursor was provided, and they suggested that the visual feedback signal allows for sensory prediction updating, whereas the reward signal allows for reward prediction updating. Motor performance adapted in both cases, but the adaptation generalized to other regions of the workspace more after the sensory updating. In the saccade domain, recent work by Madelain, Paeye, and Wallman (2011) has shown that saccades, too, can be adapted by reward information alone. In that study participants received a rewarding auditory or rewarding visual signal at the end of saccades that were of the appropriate length, instead of the retinal error that would be experienced in traditional saccadic adaptation designs. In order for the reward signal explanation to apply to our study, the participant would presumably make a comparison between the proprioceptive estimate of the hand and the visual estimate of the target, conclude that they are aligned, and use this information to reinforce the reach behaviour from that trial. While we consider this a valid possibility, there are some key differences between our study and the studies of Izawa and Shadmehr (2011) and Madelain et al. (2011) that lead us to favour an online error signal as the relevant information for learning in our study. First, unlike Izawa and Shadmehr (2011) and Madelain et al. (2011), our study did not provide any explicit reward at the completion of the movement; i.e., participants received no knowledge of results in our study and, accordingly, any reinforcing reward signal would have to have been internally generated. Second, our study involved an online response to the target perturbation, which indicates that a spatial online error signal was present and used by the motor system in real-time, with at least the potential for being used as a training signal. Third, Izawa and Shadmehr (2011) and Madelain et al. (2011) still had an error signal for their reward conditions, insofar as there would be trials where participants failed to achieve the target trajectory (Izawa & Shadmehr, 2011) or target distance (Madelain et al., 2011) and did not receive a reward. This was not true for our study. This is an important difference, because the absence of a reward on a given trial in those other studies could serve as an impetus to alter subsequent movements until a reward was obtained. One final possibility to consider is that participants in our study used the visual information that was provided as the limb was being placed at the home position to learn about their earlier reach performance on that trial. Recall that visual information was re-introduced 750 ms after the participant initiated their return movement to the home position, which they initiated after the target had been extinguished. It is, therefore, possible that the participant could rely on memory of the target location and vision of their hand near the home position, and then combine these pieces of information with knowledge of the magnitude of the motor response used to return to the home position, ultimately inferring what the endpoint error would have been on that trial. One reason we think this is unlikely is that the delay between the end of the targeted reach and the re-introduction of vision was at least 1 s (reaction time to the end-of-trial tone, which was un-speeded,
plus the 750 ms delay before light onset), a delay magnitude that has been previously shown to be detrimental to adaptation (Kitazawa, Kohno, & Uka, 1995), especially if the hand is moving to another location when vision is re-introduced (Held, Efstathiou, & Greene, 1966). Another reason we favour our online-error hypothesis over the delayed visual feedback explanation is that the similarity between the current results and those of Cameron et al. (2011) is more simply explained if we assume a common mechanism. In both studies, error was introduced online, movements were corrected online without any visual feedback of the limb, and aftereffects were observed in the posttests. 5. Conclusion Under normal reaching conditions, when people have full vision of their hand and the target, it is not clear how much of the control and learning processes is driven by visual error and how much is driven by the non-visual error processing investigated here. When available, vision of the hand may obviate any need for and, perhaps, override internal estimates of hand position. However, when vision of the hand is occluded people are able to implicitly detect, respond to, and – as we have now shown – learn from error that arises during a movement. Acknowledgments This research was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant awarded to R. Chua. B. Cameron was supported by a Juan de la Cierva postdoctoral fellowship from the Spanish government. References Bard, C., Turrell, Y., Fleury, M., Teasdale, N., Lamarre, Y., & Martin, O. (1999). Deafferentation and pointing with visual double-step perturbations. Experimental Brain Research, 125, 410–416. Bekkering, H., Abrams, R. A., & Pratt, J. (1995). Transfer of saccadic adaptation to the manual motor system. Human Movement Science, 14, 155–164. Bridgeman, B., Hendry, D., & Stark, L. (1975). Failure to detect displacement of the visual world during saccadic eye movements. Vision Research, 15, 719–722. Cameron, B. D., Franks, I. M., Inglis, J. T., & Chua, R. (2010). Implicit motor learning from target error during explicit reach control. Experimental Brain Research, 206, 99–104. Cameron, B. D., Franks, I. M., Inglis, J. T., & Chua, R. (2011). Reach adaptation to online target error. Experimental Brain Research, 209, 171–180. Desmurget, M., & Grafton, S. (2000). Forward modeling allows feedback control for fast reaching movements. Trends in Cognitive Sciences, 4, 423–431. Deubel, H., Wolf, W., & Hauske, G. (1986). Adaptive gain control of saccadic eye movements. Human Neurobiology, 5, 245–253. Goodale, M. A., Pelisson, D., & Prablanc, C. (1986). Large adjustments in visually guided reaching do not depend on vision of the hand or perception of target displacement. Nature, 320, 748–750. Held, R., Efstathiou, A., & Greene, M. (1966). Adaptation to displaced and delayed visual feedback from the hand. Journal of Experimental Psychology, 72, 887–891. Izawa, J., & Shadmehr, R. (2011). Learning from sensory and reward prediction errors during motor adaptation. PLoS Computational Biology, 7, e1002012. Kitazawa, S., Kohno, T., & Uka, T. (1995). Effects of delayed visual information on the rate and amount of prism adaptation in the human. Journal of Neuroscience, 15, 7644–7652. Krakauer, J. W., Pine, Z. M., Ghilardi, M. F., & Ghez, C. (2000). Learning of visuomotor transformations for vectorial planning of reaching trajectories. Journal of Neuroscience, 20, 8916–8924. Lackner, J. R., & Dizio, P. (1994). Rapid adaptation to Coriolis force perturbations of arm trajectory. Journal of Neurophysiology, 72, 299–313. Lawrence, M. A. (2012). ez: Easy analysis and visualization of factorial experiments. R package version 4.1-1 (http://CRAN.R-project.org/package=ez) Madelain, L., Paeye, C., & Wallman, J. (2011). Modification of saccadic gain by reinforcement. Journal of Neurophysiology, 106, 219–232. Magescas, F., & Prablanc, C. (2006). Automatic drive of limb motor plasticity. Journal of Cognitive Neuroscience, 18, 75–83. Magescas, F., Urquizar, C., & Prablanc, C. (2009). Two modes of error processing in reaching. Experimental Brain Research, 193, 337–350. McLaughlin, S. (1967). Parametric adjustment in saccadic eye movements. Perception & Psychophysics, 2, 359–362. Prablanc, C., & Martin, O. (1992). Automatic control during hand reaching at undetected two-dimensional target displacements. Journal of Neurophysiology, 67, 455–469. R Core Team (2012). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0 (URL: http:// www.R-project.org/)
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135 Shadmehr, R., & Mussa-Ivaldi, F. A. (1994). Adaptive representation of dynamics during learning of a motor task. Journal of Neuroscience, 14, 3208–3224. Tseng, Y. W., Diedrichsen, J., Krakauer, J. W., Shadmehr, R., & Bastian, A. J. (2007). Sensory prediction errors drive cerebellum-dependent adaptation of reaching. Journal of Neurophysiology, 98, 54–62.
135
Wallman, J., & Fuchs, A. F. (1998). Saccadic gain modification: Visual error drives motor adaptation. Journal of Neurophysiology, 80, 2405–2416. Welch, R. B. (1986). Adaptation of space perception. In K. R. Boff, L. Kaufman, & J. P. Thomas (Eds.), Handbook of perception and human performance. New York: Wiley and Sons.
Contents lists available at SciVerse ScienceDirect
Acta Psychologica journal homepage: www.elsevier.com/ locate/actpsy
Adapting to target error without visual feedback Brendan D. Cameron a,⁎, Jarrod Blinch b, Alyson Plecash b, Jordan Squair b, Lauren Wou b, Romeo Chua b a b
Departament de Psicologia Basica, Universitat de Barcelona, Barcelona, Catalonia, Spain School of Kinesiology, University of British Columbia, Vancouver, Canada
a r t i c l e
i n f o
Article history: Received 16 August 2012 Received in revised form 28 January 2013 Accepted 4 March 2013 Available online 2 April 2013 PsycINFO classification: 2330 Keywords: Adaptation Reaching Saccade Online control
a b s t r a c t What information is necessary for the motor system to adapt its behaviour? Visual hand-to-target error provides salient information about reach performance, but can learning proceed without this information? We investigated adaptation to an unperceived target perturbation under visual open-loop conditions. Participants looked and reached, without any vision of their hand, to a target that jumped rightward at saccade onset (Perturbation condition) or remained stationary throughout the trial (Stationary condition). The target jump in the Perturbation condition was tied to the saccade, such that participants were unaware that it had occurred. Each type of exposure was followed by a posttest, in which participants reached to a target that disappeared at saccade onset. In the posttest, participants reached farther following exposure to the perturbation than they did following exposure to the stationary target, indicating that participants had learned from systematic exposure to the jump. These findings imply that online error induces motor learning, even when participants receive no visual information about their performance. © 2013 Elsevier B.V. All rights reserved.
1. Introduction One of the ways that we refine our movements is by comparing the endpoint of a movement to a desired endpoint. When we reach for and miss a target, for instance, a visual error signal can update commands for future reaches (Magescas & Prablanc, 2006). But what if there is no visual information about the success or failure of a reach? Can the motor system use an internal estimate of limb position to adaptively modify subsequent reaches? The present study addressed this question. We already know that visual feedback is not required for real-time motor responses to target error. If a target jumps during a reach, the unseen hand will automatically deviate toward it, even when the participant is unaware of the jump (Goodale, Pelisson, & Prablanc, 1986; Prablanc & Martin, 1992). The online correction that occurs under these conditions may be driven by a proprioceptively-derived estimate of limb position or it may be driven by a forward-model-derived estimate of limb position, wherein a copy of the motor command is used to predict the limb's future position. Arguments for the involvement of forward models in these online corrections are based on the rapidity with which appropriate corrections can occur and on the observation that a deafferented patient was able to perform such corrections (Bard et al., 1999; Desmurget & Grafton, 2000). However, it is not clear if the online error signal described above operates only in real-time, to drive the online correction, or whether ⁎ Corresponding author. Tel.: +34 93 312 5143. E-mail address: [email protected] (B.D. Cameron). 0001-6918/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.actpsy.2013.03.002
it also acts as a training signal, modifying subsequent motor performance. Recent work from our group showed that unseen online target jumps produced motor adaptation, even though the online corrections eliminated terminal error (Cameron, Franks, Inglis, & Chua, 2011). In that study, we exposed participants to repeated rightward target jumps during reaches with an unseen hand, and we observed adaptation to the perturbation (cf. Magescas, Urquizar, & Prablanc, 2009). However, while vision of the hand was not available during the reach in the study of Cameron et al. (2011), vision was reintroduced at the end of each reach. So, although online correction of the reach eliminated terminal error between the hand and the target, there was visual confirmation that the target had been acquired. The present study tested for adaptation when neither online nor terminal visual feedback was available. That is, we tested whether people developed reach aftereffects after aiming to a target that imperceptibly jumped during their movement, in the absence of any real-time or terminal visual feedback. Such stimulus conditions are reminiscent of those used to induce saccadic adaptation, where the gain of saccades is adaptively altered by systematically changing the location of the saccade target while the eyes are moving (e.g., Deubel, Wolf, & Hauske, 1986; McLaughlin, 1967; Wallman & Fuchs, 1998). When the initial saccade is complete, a different distance is present between the foveated location and the location of the target than was anticipated at the start of the saccade. After repeated exposure to such error, primary saccades gradually become larger or smaller, depending on the direction of the perturbation. A key difference between motion of the eyes and the hand, however, is that the former are capable of very little, if any, online control, whereas
130
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
the latter is capable of online corrections that can eliminate some or all of the initial error in the movement. Indeed, Magescas and Prablanc's (2006) original study demonstrating adaptation of the hand to target perturbations had a protocol designed precisely to prevent any online corrections to the reach; the target was presented at its perturbed location only after completion of the reach, mimicking the error signal that is present during saccadic adaptation. Magescas and Prablanc (2006) demonstrated robust adaptation of reaching movements under such conditions, showing that terminal visual error – in the absence of any visual perturbation of the effector – was sufficient to drive adaptation. In the present study we considered a contrasting set of conditions, where online correction was possible and no visual error signal was available, to investigate whether the error signal that drives real-time reach corrections might also produce adaptive changes in the motor system. We emphasize that the present study was designed to test the role of an online error signal in the training of subsequent movements. It did not allow us to determine whether the nature of the error signal was proprioceptive and/or efference copy-based, and we made no assumptions about which of these signals was potentially driving adaptation. 2. Materials and methods 2.1. Participants Eight participants from the university community (4 male, 4 female; ages 20–22) completed the study. All participants were self-described right-handed, had normal or corrected-to-normal vision, and were naïve to the aims of the study. All participants provided informed consent prior to the experiment, and the study was conducted in accordance with the guidelines of the university's research ethics board. 2.2. Apparatus Stimuli were projected onto a half-silvered mirror mounted midway between a horizontal reaching surface and an inverted LED array, such that targets (LEDs) appeared to be in the same plane as the reaching hand. Participants rested their head in a chin-rest, such that their eyes were 50 cm from the reaching surface. Vision of the hand was occluded when a white light below the mirror was extinguished. The apparatus was located in a dark room. Electrooculography (EOG) was used to monitor horizontal saccades. Disposable Ag–AgCl surface electrodes were placed at the outer canthi of the eyes with a reference electrode on the forehead. EOG signals were amplified (5–10 K) and band-pass filtered (0.1–30 Hz) with an AC preamplifier (Grass Instruments P511). The experimenter manually set a voltage threshold for each participant such that the trigger would occur within approximately the first half of the saccade. The first peak in the EOG signal was interpreted as completion of the primary saccade. The experimenter used an onscreen display of the EOG signal to visually estimate the magnitude of the participant's saccades and then set the voltage threshold accordingly. Precise calculation of saccade magnitude was done offline. Hand movements were recorded with Optotrak (Northern Digital), sampling at 500 Hz. An infrared emitting diode was mounted on the tip of a hand-held stylus. The stylus was also equipped with a pressure sensitive tip that was used to record movement onset and end. The experiment was run with a dedicated computer running DOS, with a digital I/O card with high-resolution timers (10 kHz). For control of the target jump within a trial, the analog EOG signal was fed through an independent hardware analog circuit that, when the EOG signal exceeded a set threshold, set a transistor–transistor logic trigger. This trigger (square wave) signal was used to trigger the LED onset/offset for the target jump. The delay between this trigger being set and the LED being triggered was less than 1 ms.
2.3. Task Participants began each trial with the stylus placed at the home position and their eyes on the fixation point, 18.5 cm above the home position. After a variable foreperiod a target would appear 21 cm to the right of fixation (Center location), coincident with the offset of fixation. On look-and-reach trials (indicated by a red fixation point), the participant's task was to look and reach to the target in a single, smooth, and accurate motion. If movement time was faster than 300 ms or slower than 600 ms, participants were asked to slow down or speed up on the next trial, accordingly. On look-only trials (indicated by a green fixation point), the participant's task was to only look at the target, keeping their hand at the home position. Look-only trials were interleaved with reaching trials in the Practice and Exposure phases of the experiment, described next. 2.4. Conditions and phases Each participant completed two conditions on the same day, an experimental condition (Perturbation) and a control condition (Stationary), the order of which was counterbalanced across participants. Each condition consisted of 4 phases: Practice (12 look-and-reach trials, 12 look-only trials), Exposure (35 look-and-reach, 35 look-only), Posttest 1 (20 look-and-reach), and Posttest 2 (10 look-and-reach), in that order. In the Practice phase, participants received full vision of their hand and the target throughout each trial. In all other phases, vision of the hand was available at the start of the trial, but then extinguished for the entirety of the reach (lights off at saccade start). One second after completion of the reach, the target was extinguished (if it was present) and a tone was sounded, indicating to participants that they could return their hand to the home position. Then, 750 ms after the participant began to return to the home position, vision of the hand was re-introduced to allow precise placement of the stylus at the home position. The only difference between the Perturbation and Stationary conditions occurred in the Exposure phase. In the Perturbation condition, the target jumped during the Exposure phase: 4.7 cm right (from Center to Right location) on look-and-reach trials and 4.7 cm left (from Center to Left location) on look-only trials. We included lookonly trials in order to inhibit the accumulation of any saccadic adaptation on look-and-reach trials (Cameron et al., 2011; Magescas et al., 2009). If such adaptation occurred, it might transfer to the reaching limb (Bekkering, Abrams, & Pratt, 1995). The target jump, which consisted of extinguishing an LED at one location and immediately illuminating an LED at another location, was triggered during the saccade to eliminate any awareness of the jump (Bridgeman, Hendry, & Stark, 1975). In the Stationary condition, the target did not jump during the Exposure phase. Instead, it initially appeared at the Right location and remained at that location throughout the reach (look-and-reach trials) or initially appeared at the Left location and remained at that location throughout the eye-movement (look-only trials). On Posttest 1 trials the target appeared at the Center location, then disappeared at saccade start and did not reappear. This first posttest, in which no target was present during the reach, was designed to detect any changes in movement planning that resulted from exposure to the perturbation. If adaptation was to accumulate, it should manifest in Posttest 1 as positive endpoint bias in the Perturbation condition relative to the Stationary condition. This contrasts with Posttest 2 trials, in which the target appeared at the Center location and remained lit until 1 s after movement completion. This second posttest, in which a stationary target was visible during the reach, was designed to wash out any accumulated adaptation. Here, we would expect online correction to the stationary target and an unlearning of any adaptation that may have occurred during exposure. Our protocol was designed for a comparison between the Perturbation and Stationary conditions in the posttest phases, rather than a
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
comparison between a pretest and posttest phase in the Perturbation group. We deliberately excluded pretests from our design. In prior experiments where we extinguished hand and target vision during reaching we observed sizeable drifts in movement endpoint over the course of reaching: participants increasingly undershot the target as trials progressed (Cameron, Franks, Inglis, & Chua, 2010; Cameron et al., 2011). Accordingly, had we included a pretest in the current study, we would have expected drift to begin in that phase and continue to accumulate throughout subsequent phases. With recalibration not possible during the no-vision exposure condition, we would expect drift-driven undershooting to be larger in the posttest than in the pretest, thus masking some or all of the adaptation to the target error. In the current design the Stationary and Perturbation conditions should accumulate equivalent amounts of endpoint drift, and comparison between them should consequently reveal any target-driven adaptation in the Perturbation condition.
131
reach onset – a measure intended to index the prepared response – no significant effects emerged (p > .10). It is not clear whether an absence of a significant perturbation effect at this early time point suggests that the aftereffect did not extend to motor preparation (and, rather, was restricted to online control), or whether the effect of the perturbation on motor preparation was simply too small to detect at such an early stage of the reach. Because the visual target was extinguished at reach onset in Posttest 1, we favour the view that performance in that phase reflected a prepared response. The velocity profiles in the early posttest (Section 3.3), which suggest a more robust response in the perturbation condition, also support this interpretation; however, we cannot rule out the possibility that some or all of the effects we have observed result from an influence of the perturbation on online control only. 3.2. Real-time response to the perturbation: performance in the Exposure phase
2.5. Analysis To assess the effects of the perturbation we examined radial position of the hand relative to the home position. We examined the position of the hand at 3 time points: reach endpoint, 200 ms after reach onset, and 100 ms after reach onset, the last of which was intended as an index of the prepared movement. For each participant, position averages were calculated for every “block” of 5 consecutive trials. For the Exposure phase we examined the hand's position at each time point (reach endpoint, 200 ms, and 100 ms) with separate 2 Condition × 7 Block repeated measures ANOVAs. For Posttest 1 we applied 2 Condition × 4 Block repeated measures ANOVAs for each time point. The Huynh–Feldt correction to degrees of freedom was applied in the case of sphericity violations, which were detected with Mauchley's at p b .10. Statistical analyses were carried out with the ‘ez’ package (Lawrence, 2012) in R (R Core Team, 2012). We obtained the magnitude of the primary saccade on each trial by first locating, in the EOG trace, the first positive voltage peak after the onset of the target. We worked backwards from this peak to locate the first local minimum (the onset of the saccade). The voltage difference between these two points was used as a measure of saccade magnitude. For statistical analysis, participants' mean voltage values were submitted to a 2 Condition × 7 Block repeated measures ANOVA in the Exposure phase and a 2 Condition × 4 Block repeated measures ANOVA in Posttest 1. 3. Results and discussion 3.1. Adaptation to the perturbation: performance in the Posttest 1 phase To determine if people adapted to the perturbation, we first compared movement endpoints from the Posttest 1 phases of the Perturbation and Stationary conditions. If the motor system implicitly learned from exposure to the target jump, we should observe farther reach endpoints in the posttest of the Perturbation condition than in the posttest of the Stationary condition. As shown in Fig. 1 (top), participants reached farther after exposure to the perturbation than after exposure to a stationary target, F(1,7) = 32.78, p b .001. We also observed a condition × block interaction, F(3,21) = 3.35, p = .04, indicating that the endpoint difference between conditions was not constant across all blocks of the posttest. The size of the difference decreased after the first block of the posttest (Fig. 1, top). Earlier stages in the reach also showed a pattern of farther reaching in Posttest 1 for the perturbation condition than for the stationary condition (Fig. 1, middle and bottom panels). At 200 ms after reach onset there was a significant effect of condition, F(1,7) = 7.87, p b .05, and a marginally significant interaction, F(3,21) = 2.91, p = .058, suggesting, like the endpoint results, that the aftereffect of the perturbation was strongest early in the posttest. At 100 ms after
Analysis of endpoint performance in the Exposure phase suggested that participants were largely effective at responding online to the target jump in the Perturbation condition (Fig. 1). Endpoints did not differ significantly between the two conditions in the Exposure phase, F(1,7) b 1. Recall that the target jumped from the Center to the Right location in the Perturbation condition, whereas it was always present at the Right location in the Stationary condition. So, if participants had failed to respond online to the jump, a relatively large difference (~3.7 cm in radial distance) would have been present between the endpoints for the two conditions in the Exposure phase. Furthermore, when the 35 look-and-reach trials of the Exposure phase were considered as 7 consecutive blocks of 5 trials, there was no interaction between block and condition, F(6,42) b 1. However, an absence of a significant difference here does not necessarily mean that the endpoints were equivalent between conditions. Some undershooting may have been present in the perturbation condition that we were not able to detect statistically. Although the F-value for the condition comparison was very small (less than 1) for the radial distance measure, it was larger for the lateral position measure (condition effect: F(1,7) = 3.11, p = .12, data not shown). In other words, there may have been a trend in the Exposure phase for more leftward endpoint error in the Perturbation condition than in the Stationary condition, and we cannot rule this out as a possible contributor to the adaptation. Radial distance at 100 ms after reach onset is a measure that should be free from contamination by online processing. Accordingly, distance at 100 ms, tracked across blocks of trials, should tell us about changes in movement preparation in the Exposure phase caused by repeated perturbations (Fig. 1, bottom). We anticipated a constant distance across blocks for the Stationary condition (for which the target was at the same location from the start and end of each trial) and a gradually increasing distance across blocks for the Perturbation condition. Such an effect would manifest as an interaction between the condition and block factors. However, we observed only an effect of block, F(2.6,17.9) = 5.56, p b .01, indicating a decrease in distance travelled for both conditions over blocks. Neither the interaction effect, F(6,42) = 1.99, p = .09, nor the condition effect, F(1,7) b 1 was significant. The absence of the predicted interaction effect at 100 ms may indicate, as discussed in the previous section, that insufficient time had passed for the effect to emerge. At 200 ms, a significant interaction effect was present, F(6,42) = 2.53, p b .05, though it is unclear whether its pattern (Fig. 1, middle) is consistent with what we had predicted. There was also a significant effect for block, F(6,42) = 5.34, p b .001, but no effect for condition, F(1,7) b 1. To summarize: the endpoint analysis of the Exposure phase suggested that participants corrected their movements online to the perturbation, reaching endpoint locations that were comparable, if not necessarily equivalent to those reached when the target was
132
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
320 280
300
Stationary Perturbation
260
Radial distance from start (mm)
Reach endpoint (mean MT = 394 ms)
Exposure
Post1
Post2
180
200
220
Stationary Perturbation
160
Radial distance from start (mm)
200 ms after reach onset
Exposure
Post1
Post2
80 50
60
70
Stationary Perturbation
40
Radial distance from start (mm)
100 ms after reach onset
Exposure
Post1
Post2
Fig. 1. Radial distance (relative to the home position) travelled by the stylus at reach endpoint (top panel), 200 ms after reach onset (middle panel), and 100 ms after reach onset (bottom panel). Each data point represents the mean of 5 consecutive trials. In the top panel, the upper horizontal dashed line indicates the radial distance of the Right target location, while the lower horizontal dashed line indicates the radial distance of the Center target location. Error bars denote 1 standard error of the mean. Note the different scaling of the y-axis for the bottom panel.
stationary at the farther location. The analysis of distance travelled at 100 ms and 200 ms provided ambiguous results. There was some indication from the interaction effect for the 200 ms measure that distance travelled increased for the Perturbation condition relative to the Stationary condition, but the pattern was not clear, and we accordingly refrain from claiming that an adaptation effect was observable within the Exposure phase (though we do think the presence of adaptation can be clearly inferred from the results of the posttest described in the previous section). 3.3. Velocity profiles in the Exposure and Posttest 1 phases We examined the velocity profiles at the start (first 5 trials) and end (last 5 trials) of the Exposure and Posttest 1 phases (Fig. 2). These profiles suggest comparable performance for Stationary and Perturbation conditions in the Exposure phase, but larger amplitude movements for the Perturbation condition than the Stationary condition in the early Posttest 1 phase. Analysis of the peak velocities at each of these stages revealed a significantly larger peak velocity for the Perturbation condition in early Posttest 1, t(7) = 3.97, p b .01, but no significant differences between the conditions in early Exposure, t(7) = −.78,
p > .20, late Exposure, t(7) = 1.35, p > .20, or late Posttest 1, t(7) = 1.66, p > .10.
3.4. Saccadic adaptation was not responsible for reach adaptation Primary saccade amplitudes in Posttest 1 did not differ significantly between the Stationary condition (mean: 2.5 V) and the Perturbation condition (mean: 2.3 V), F(1,7) = 1.37, p = .28. This implies that the opposite direction target jumps on the saccade-only trials during the Exposure phase successfully inhibited any accumulation of rightward saccadic adaptation, consistent with previous studies that have used this method (Cameron et al., 2011; Magescas et al., 2009). Accordingly, any difference in reach performance between the conditions is not likely to be due to saccade-related processes. That is, farther reaching in the Perturbation condition was not a consequence of saccade magnitude. In the Exposure phase, by contrast, primary saccade amplitude was larger in the Stationary condition than in the Perturbation condition, F(1,7) = 32.06, p b .001. This was expected, as the initial target location was 4.7 cm farther right in the look-and-reach trials of the Stationary Exposure phase, and it serves as confirmation that our EOG
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
Exposure End 1000 600
Velocity (mm/s)
600 200
Stationary Perturbation
0
200
1000
Stationary Perturbation
0
Velocity (mm/s)
Exposure Start
0
100
200
300
400
0
Time after reach onset (ms)
100
200
300
400
Time after reach onset (ms)
Post1 End
600
Velocity (mm/s)
600 200
0
0
Stationary Perturbation
200
1000
Stationary Perturbation
1000
Post1 Start
Velocity (mm/s)
133
0
100
200
300
400
Time after reach onset (ms)
0
100
200
300
400
Time after reach onset (ms)
Fig. 2. Mean resultant velocity profiles at 4 time points: the first 5 trials (Start) and last 5 trials (End) of the Exposure and Posttest 1 phases. To obtain the displayed means, trajectories were locked to movement onset, cropped at 400 ms, and collapsed across all participants.
measure of saccade magnitude was sensitive to target location, as we have shown previously (Cameron et al., 2011). 3.5. Participants were unaware of the target jump Each participant was asked at the end of the experiment whether they saw the target jump. Two participants thought they might have seen the target jump one time during the experiment. (It jumped 70 times). All other participants reported no awareness of the jump. Furthermore, when asked whether they noticed a difference between the two sessions of the experiment (i.e., Stationary vs. Perturbation), all participants reported no difference. This lack of awareness of the jump is consistent with previous studies employing comparable saccadically-triggered target jumps (e.g., Bridgeman et al., 1975; Cameron et al., 2011; Goodale et al., 1986; Magescas et al., 2009; Prablanc & Martin, 1992). 3.6. Timing of the target jump with respect to the eye and hand Here we report the relative timing of the eye, hand, and perturbed target onsets in the Exposure phase of the Perturbation condition. (Group means are presented, followed by mean within-subject standard deviations in brackets.) When participants looked and reached for the target, the eye preceded the hand by 93.6 ms (73.7), and the target jump was triggered 33 ms (6.3) after the onset of the saccade. Mean saccade duration in this condition was 76.3 ms (7.7), and mean movement time of the reach was 385.8 ms (18.7). 4. General discussion In our experiment, people adapted to an unperceived target jump without any visual feedback. This suggests that error that arises during a reach – error that the motor system compensates for in real-time – informs performance on subsequent reaches. Moreover, it suggests
that learning occurs even when this error signal relies on non-visual estimates of effector position. Motor adaptation studies have traditionally manipulated either the force acting on the limb, via a robotic arm (e.g., Shadmehr & MussaIvaldi, 1994) or Coriolis forces (e.g., Lackner & Dizio, 1994), or the visual feedback of the limb, via prisms (e.g., Welch, 1986) or cursor rotations (e.g., Krakauer, Pine, Ghilardi, & Ghez, 2000). The present study contrasts with such studies, as neither the force acting on the limb nor vision of the limb was altered. This is in line with work by Magescas and Prablanc (2006), who demonstrated adaptation to terminal reach error induced by a target perturbation. In that study, however, the critical adaptive signal was visual hand-to-target error. Our results suggest that reach error needs to be neither terminal nor visual in order for adaptation to target error to occur. One possibility that we have not yet addressed is that the online correction itself, rather than the error signal that drives it, is responsible for the adaptation that we observed. While the present study cannot distinguish between these two possibilities, previous studies of reaches (Tseng, Diedrichsen, Krakauer, Shadmehr, & Bastian, 2007) and of saccades (Wallman & Fuchs, 1998) suggest that it is the error signal, rather than the corrective motor response, that is primarily responsible for adaptation. We, accordingly, favour the argument that the real-time error signal, rather than the motor correction, was responsible for the adaptation observed in the present study. However, the studies of Tseng et al. and Wallman and Fuchs tested a visual error signal (absent in the present study), and so the applicability of their findings to the present study is not clear. A future study applying the present study's visual manipulations while testing both rapid, uncorrectable movements and slower, correctable ones (a manipulation employed by Tseng et al., 2007) might address this question. Though we have suggested that our results indicate an effect of online error (and/or online correction) on reach adaptation, an alternative explanation is that terminal feedback provided the critical signal for adaptation. Although participants in our study did not receive any visual feedback at the end of their reach, proprioceptive feedback was always
134
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135
available, and it may have permitted a comparison between the visual target and the hand's location at the end of the reach. Movement endpoints did not differ significantly between the Perturbation and Stationary conditions during the Exposure phase, and we took this as evidence that terminal error signals were comparable between the two conditions, arguing in favour of the online error signal in the Perturbation condition being the critical difference. However, our argument works only if one considers a spatial error signal to be the sole driver of motor adaptation. A terminal reward signal, rather than a spatial error signal, could be responsible for the effects we observed. Izawa and Shadmehr (2011) recently showed that a reward signal, in the absence of any visual feedback, is able to produce adaptation of reaching movements. In the reward condition of that study, participants were only provided with an indication of success or failure of the reach (the target exploded if the reach was successful). They compared this to conditions where visual feedback of a rotated cursor was provided, and they suggested that the visual feedback signal allows for sensory prediction updating, whereas the reward signal allows for reward prediction updating. Motor performance adapted in both cases, but the adaptation generalized to other regions of the workspace more after the sensory updating. In the saccade domain, recent work by Madelain, Paeye, and Wallman (2011) has shown that saccades, too, can be adapted by reward information alone. In that study participants received a rewarding auditory or rewarding visual signal at the end of saccades that were of the appropriate length, instead of the retinal error that would be experienced in traditional saccadic adaptation designs. In order for the reward signal explanation to apply to our study, the participant would presumably make a comparison between the proprioceptive estimate of the hand and the visual estimate of the target, conclude that they are aligned, and use this information to reinforce the reach behaviour from that trial. While we consider this a valid possibility, there are some key differences between our study and the studies of Izawa and Shadmehr (2011) and Madelain et al. (2011) that lead us to favour an online error signal as the relevant information for learning in our study. First, unlike Izawa and Shadmehr (2011) and Madelain et al. (2011), our study did not provide any explicit reward at the completion of the movement; i.e., participants received no knowledge of results in our study and, accordingly, any reinforcing reward signal would have to have been internally generated. Second, our study involved an online response to the target perturbation, which indicates that a spatial online error signal was present and used by the motor system in real-time, with at least the potential for being used as a training signal. Third, Izawa and Shadmehr (2011) and Madelain et al. (2011) still had an error signal for their reward conditions, insofar as there would be trials where participants failed to achieve the target trajectory (Izawa & Shadmehr, 2011) or target distance (Madelain et al., 2011) and did not receive a reward. This was not true for our study. This is an important difference, because the absence of a reward on a given trial in those other studies could serve as an impetus to alter subsequent movements until a reward was obtained. One final possibility to consider is that participants in our study used the visual information that was provided as the limb was being placed at the home position to learn about their earlier reach performance on that trial. Recall that visual information was re-introduced 750 ms after the participant initiated their return movement to the home position, which they initiated after the target had been extinguished. It is, therefore, possible that the participant could rely on memory of the target location and vision of their hand near the home position, and then combine these pieces of information with knowledge of the magnitude of the motor response used to return to the home position, ultimately inferring what the endpoint error would have been on that trial. One reason we think this is unlikely is that the delay between the end of the targeted reach and the re-introduction of vision was at least 1 s (reaction time to the end-of-trial tone, which was un-speeded,
plus the 750 ms delay before light onset), a delay magnitude that has been previously shown to be detrimental to adaptation (Kitazawa, Kohno, & Uka, 1995), especially if the hand is moving to another location when vision is re-introduced (Held, Efstathiou, & Greene, 1966). Another reason we favour our online-error hypothesis over the delayed visual feedback explanation is that the similarity between the current results and those of Cameron et al. (2011) is more simply explained if we assume a common mechanism. In both studies, error was introduced online, movements were corrected online without any visual feedback of the limb, and aftereffects were observed in the posttests. 5. Conclusion Under normal reaching conditions, when people have full vision of their hand and the target, it is not clear how much of the control and learning processes is driven by visual error and how much is driven by the non-visual error processing investigated here. When available, vision of the hand may obviate any need for and, perhaps, override internal estimates of hand position. However, when vision of the hand is occluded people are able to implicitly detect, respond to, and – as we have now shown – learn from error that arises during a movement. Acknowledgments This research was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant awarded to R. Chua. B. Cameron was supported by a Juan de la Cierva postdoctoral fellowship from the Spanish government. References Bard, C., Turrell, Y., Fleury, M., Teasdale, N., Lamarre, Y., & Martin, O. (1999). Deafferentation and pointing with visual double-step perturbations. Experimental Brain Research, 125, 410–416. Bekkering, H., Abrams, R. A., & Pratt, J. (1995). Transfer of saccadic adaptation to the manual motor system. Human Movement Science, 14, 155–164. Bridgeman, B., Hendry, D., & Stark, L. (1975). Failure to detect displacement of the visual world during saccadic eye movements. Vision Research, 15, 719–722. Cameron, B. D., Franks, I. M., Inglis, J. T., & Chua, R. (2010). Implicit motor learning from target error during explicit reach control. Experimental Brain Research, 206, 99–104. Cameron, B. D., Franks, I. M., Inglis, J. T., & Chua, R. (2011). Reach adaptation to online target error. Experimental Brain Research, 209, 171–180. Desmurget, M., & Grafton, S. (2000). Forward modeling allows feedback control for fast reaching movements. Trends in Cognitive Sciences, 4, 423–431. Deubel, H., Wolf, W., & Hauske, G. (1986). Adaptive gain control of saccadic eye movements. Human Neurobiology, 5, 245–253. Goodale, M. A., Pelisson, D., & Prablanc, C. (1986). Large adjustments in visually guided reaching do not depend on vision of the hand or perception of target displacement. Nature, 320, 748–750. Held, R., Efstathiou, A., & Greene, M. (1966). Adaptation to displaced and delayed visual feedback from the hand. Journal of Experimental Psychology, 72, 887–891. Izawa, J., & Shadmehr, R. (2011). Learning from sensory and reward prediction errors during motor adaptation. PLoS Computational Biology, 7, e1002012. Kitazawa, S., Kohno, T., & Uka, T. (1995). Effects of delayed visual information on the rate and amount of prism adaptation in the human. Journal of Neuroscience, 15, 7644–7652. Krakauer, J. W., Pine, Z. M., Ghilardi, M. F., & Ghez, C. (2000). Learning of visuomotor transformations for vectorial planning of reaching trajectories. Journal of Neuroscience, 20, 8916–8924. Lackner, J. R., & Dizio, P. (1994). Rapid adaptation to Coriolis force perturbations of arm trajectory. Journal of Neurophysiology, 72, 299–313. Lawrence, M. A. (2012). ez: Easy analysis and visualization of factorial experiments. R package version 4.1-1 (http://CRAN.R-project.org/package=ez) Madelain, L., Paeye, C., & Wallman, J. (2011). Modification of saccadic gain by reinforcement. Journal of Neurophysiology, 106, 219–232. Magescas, F., & Prablanc, C. (2006). Automatic drive of limb motor plasticity. Journal of Cognitive Neuroscience, 18, 75–83. Magescas, F., Urquizar, C., & Prablanc, C. (2009). Two modes of error processing in reaching. Experimental Brain Research, 193, 337–350. McLaughlin, S. (1967). Parametric adjustment in saccadic eye movements. Perception & Psychophysics, 2, 359–362. Prablanc, C., & Martin, O. (1992). Automatic control during hand reaching at undetected two-dimensional target displacements. Journal of Neurophysiology, 67, 455–469. R Core Team (2012). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing. ISBN 3-900051-07-0 (URL: http:// www.R-project.org/)
B.D. Cameron et al. / Acta Psychologica 143 (2013) 129–135 Shadmehr, R., & Mussa-Ivaldi, F. A. (1994). Adaptive representation of dynamics during learning of a motor task. Journal of Neuroscience, 14, 3208–3224. Tseng, Y. W., Diedrichsen, J., Krakauer, J. W., Shadmehr, R., & Bastian, A. J. (2007). Sensory prediction errors drive cerebellum-dependent adaptation of reaching. Journal of Neurophysiology, 98, 54–62.
135
Wallman, J., & Fuchs, A. F. (1998). Saccadic gain modification: Visual error drives motor adaptation. Journal of Neurophysiology, 80, 2405–2416. Welch, R. B. (1986). Adaptation of space perception. In K. R. Boff, L. Kaufman, & J. P. Thomas (Eds.), Handbook of perception and human performance. New York: Wiley and Sons.