Looking at the task in hand impairs motor learning

Abstract

“Visual capture” is the term used to describe vision being afforded a higher weighting than other sensory information. Visual capture can produce powerful illusory effects with individuals misjudging the size and position of their hands. The advent of laparoscopic surgical techniques raises the question of whether visual capture can interfere with an individual's rate of motor learning. We compared adaptation to distorted visual feedback in two groups: the Direct group appeared to have the advantage of directly viewing the input device, while the Indirect group used the same input device but viewed their movements on a remote screen. Counterintuitively, the Indirect group adapted more readily to distorted feedback and showed enhanced performance. The results show that visual capture impairs adaptation to distorted visual feedback, suggesting that surgeons need to avoid viewing their hands when learning laparoscopic techniques.

tastevin first coined the term capitation visuelle to describe the modification of haptic perception by vision (Tastevin 1937). Visual capture can give rise to remarkable phenomena. Participants report square objects as rectangular when subjected to optical distortion despite veridical haptic information (Rock and Victor 1964). Similarly, individuals will erroneously point an unseen finger to the displaced visual location of a finger on the other hand (Mon-Williams et al. 1997b). Visual capture can decrease phantom sensation in amputated arms (Ramachandran et al. 1995), and if participants see a model hand instead of their own hand then they report feeling mechanical forces applied to the fake hand: the “rubber hand” illusion (Botvinick and Cohen 1998). The rubber hand illusion is strong enough to elicit pain if the fake hand is subjected to pin pricks (Capelari et al. 2009). Visual capture likewise affects tactile discrimination, with impaired performance occurring when visual distracters are congruent with tactile stimuli (Pavani et al. 2000).

Visual capture is remarkable at a phenomenological level but also provides insights into nervous system organization. The phenomenon indicates that information is used from a variety of sources and is integrated to generate a stable and consistent representation of the world. The fact that vision can “dominate” other cues shows that information is weighted according to its reliability. Indeed, Mon-Williams and colleagues (1997b) showed that when pointing an unseen finger toward their visible hand, participants relied more upon the felt location when vision was degraded. It has been found that visual and haptic information are combined in an optimal manner that accounts for the noise associated with the signals (Hillis et al. 2002; van Beers et al. 1999).

In a stable physical world the combination of information from different sources has a clear evolutionary advantage for a nervous system learning how to interact with the environment (Landy et al. 1995), but recent forms of technology can create a mismatch between perception and action. For example, humans are remarkably adept at maneuvering a visual cursor around a computer screen via a spatially distant handheld input device (e.g., a computer mouse). It is a matter of common observation that humans adapt rapidly to changes in the dynamics of an input device and show outstanding proficiency in learning novel input-output relationships (Wolpert and Flanagan 2010). The advent of laparoscopic (and robotic) surgical devices placed particular demands on the learning of a novel input-output map. Laparoscopic devices require surgeons to move manipulanda through complex force fields in order to achieve a specific surgical outcome (Wagner et al. 1995). The complexity of the task is increased because online feedback about the progress of the movements is presented remotely (typically the images from a camera inside the patient's body are displayed on a screen), with nonlinear distortions created between the direction of movement of the laparoscopic device and the resultant movement of the image on the two-dimensional screen (Wagner et al. 1995).

In the present experiment we were interested in exploring whether visual capture interferes with learning a novel input-output map. Intuitively, it seems that fixating the hands when learning laparoscopic surgery would provide additional information that might aid skill acquisition. Conversely, previous findings regarding visual capture suggest that such visual information might actually interfere with the learning processes.

METHODS

Forty-four healthy adults participated in the University of Leeds ethics committee-approved study after giving written informed consent (age range 19–31 yr, mean 23.45 yr, SD = 3.08 years; 39 right-handed, 5 left-handed participants). All participants reported normal or corrected-to-normal eyesight and no history of neurological problems. Participants were allocated to one of four experimental groups with 11 participants in each (see Fig. 1). Participants sat comfortably in an adjustable-height chair in front of a Toshiba Portege M700 Series tablet laptop (screen: 260 × 163 mm, 1,440 × 900 pixels, 32-bit color, and 60-Hz refresh rate) laid out in the tablet position. This was used as the input surface for the tasks, and a digitizing stylus was used as the input device (see Fig. 1). The laptop ran specialized software to present the stimuli and record kinematic measures at 100 Hz, capturing the x and y coordinates of the stylus position (Culmer et al. 2009). Standardized audible and written instructions were delivered by the computer at the start of each block. Two visual feedback groups were used: 1) Direct feedback, where the participants could see their hand and stylus on the tablet screen (Fig. 1A), and 2) Indirect feedback, where the stylus and hand were hidden from view (Fig. 1B). A specially made shelf was placed over the tablet laptop on the desk for the Indirect feedback group, with a screen obscuring view of the tablet laptop and the hand and arm. An external monitor was placed on top of the shelf and connected to the tablet laptop in order to display the visual stimuli. Participants' eyes were approximately level with the center of the monitor, and participants sat at a distance of ∼46 cm from the monitor (the head was not restrained). This setup is similar to the conditions experienced by laparoscopic surgeons, who usually view a camera image of the surgical site on a remote display positioned in a different plane from the coordinate frame where action takes place.

Fig. 1.

The task required participants to move a circular marker to a series of targets. When the marker touched 1 target, the next target in the sequence appeared (the whole sequence is shown here for clarity). A straight line was drawn between the targets to provide a reference path. Two visual feedback groups were used: Direct feedback, where the participants could see their hand and stylus (A), and Indirect feedback, where the stylus and hand were hidden from view (B). All participants completed 4 experimental blocks over 2 consecutive days: 3 blocks on day 1 and 1 block on day 2. Each block consisted of 15 individual trials. On day 1, baseline performance was recorded first with no distortion, and then training was carried out with 2 distortions: 30° (C) and 60° (D). The distortions introduced a rotation that meant there was a mismatch between the input device and the circular marker. Each distortion was completed as a block, but half of each feedback group experienced the small then large distortion (30°/60°) and the other half the large then small distortion (60°/30°). A short break was provided between blocks. On day 2 a “transfer” block was completed with 45° rotation. E: examples from a representative participant during the first 60° trial (top), the final 60° trial (middle), and a baseline trial without distortion (bottom). The first trial clearly shows a series of erroneous “wiggly” movements that are repeatedly corrected, but by the 15th trial the hand is moving almost as fluently as at baseline.

Procedures.

The task required participants to move a circular marker to a series of 19 targets (the first movement from the start to the first target and the final movement from the last target to the finish were excluded). When the marker touched one target, the next target in the sequence appeared (the whole sequence is shown throughout Fig. 1). Targets were on average 20 mm apart (ranging between 18 and 22 mm), and a line was drawn between them to provide a reference for the straight-line path. The same sequence was used in every trial in order to ensure that movement requirements were identical (e.g., number and distribution of movement directions and angles), which meant that measurements such as movement time (MT) were comparable throughout the experiment (so a reduction in MT clearly reflected quicker movements). A sequence order was chosen that required the participant to move and explore the mapping across a broad extent of the workspace. One limitation was that when the mapping was rotated by 60° it still had to be possible to move the hand on the tablet surface so the circular marker could reach the position of each target (this was confirmed by pilot testing).

All participants completed four experimental blocks over two consecutive days: three blocks on day 1 and one block on day 2. Each block consisted of 15 individual trials. On day 1, baseline performance was recorded first with no distortion, and then training was carried out with two distortions: 1) 30° and 2) 60°. The distortions introduced a rotation that meant there was a mismatch between the input device and the circular marker. Each distortion was completed as a block, but half of each feedback group experienced the small then large distortion (30°/60°) and the other half the large then small distortion (60°/30°). A short break was provided between blocks. On day 2 a “transfer” block was completed with a 45° rotation.

MT was taken as the primary performance measure since end-point accuracy was a necessary requirement to complete the task. We also calculated the path length (PL) between targets to measure how straight the trajectories were. To generate a simple measure of learning we calculated an MT and PL measure at the beginning and end of each block. Specifically, the first five trials of each block were averaged for each participant to give a measure of initial performance, and the last five trials of each block were also averaged to give a final performance measure for each participant. Hence, eight scores were obtained for each participant for both MT and PL.

Movement time analysis.

First, to establish the baseline performance, the MT for Direct and Indirect feedback groups were compared for the 0° rotation condition. A mixed-model ANOVA with a 2 (group: Direct, Indirect) × 2 (trials: first 5, last 5) design was used. Figure 3 shows that learning rates had reduced considerably by the final five trials (actual rates were calculated and are shown in Table 5) and so provided a useful baseline for our purposes. Average performance across the last five trials for each person was used as a baseline and was subtracted from that individual's performance during training and at transfer. The rationale for this subtraction is that there may well be performance differences between the groups unrelated to the distortions applied during training and at transfer (e.g., the different viewing plane or individual differences) and we principally wanted to examine how effectively each group learns relative to their baseline performance.

Fig. 3.

Movement times averaged across each trial for Direct (A and C) and Indirect (B and D) viewing groups for the 2 condition orders (60°-30° or 30°-60°). Rotation conditions were carried out as follows: 0° was the baseline performed on day 1, followed by either 60° then 30° (A and B) or 30° then 60° (C and D); finally, 45° was performed on day 2.

Table 5.

Average straight-line gradient fitted to first five and last five trials of each participant for each of the rotation conditions in Direct or Indirect viewing group

	Direct		Indirect			Direct		Indirect
Condition Order	F5	L5	F5	L5	Condition Order	F5	L5	F5	L5
0°	−1.55	−0.15	−2.28	−0.25	0°	−1.22	−0.06	−2.27	−0.24
30°	−1.67	−0.20	−1.97	0.02	60°	−9.48	−0.52	−3.29	−0.27
60°	−1.62	−0.29	−1.20	−0.28	30°	−0.64	−0.10	−0.72	0.03
45°	−0.86	0.07	−1.29	−0.18	45°	−1.44	−0.11	−0.77	−0.07

Gradient values indicate the reduction in MT that occurs between each trial included in the fit (for example, a gradient of −.2 indicates a reduction in MT by 1 s by the end of 5 trials). Rotation conditions were carried out as follows: 0° was the baseline performed first on day 1, followed by 30° and 60° in the order shown, then finally 45° was performed on day 2. F5, first 5 trials; L5, last 5 trials.

To determine whether performance was different for Direct and Indirect feedback groups at transfer (day 2, 45°) we performed a mixed-model ANOVA with a 2 (group: Direct, Indirect) × 2 (trials: first 5, last 5) × 2 (training order: 30° then 60°, 60° then 30°) design. To examine whether performance was different for Direct and Indirect feedback groups during training, a further mixed-model ANOVA was performed with a 2 (group: Direct, Indirect) × 2 (trial time: first 5, last 5) × 2 [rotation size: small (30°), large (60°)] design.

Path length analysis.

Movement accuracy often provides a useful measure of performance (in addition to MT). Because in our task one target must be reached before the next target is drawn, end-point accuracy is not a discriminating measure. We were able to examine, however, how straight the trajectories were since the “ideal” (and shortest) path is a straight line. We refer to this movement accuracy measure as path length (PL). To determine whether there are group differences in PL we carried out ANOVAs similar to our examination of MT. First, we compared performance between Direct and Indirect feedback groups at baseline (0° rotation), using a mixed-model ANOVA with a 2 (group: Direct, Indirect) × 2 (trials: first 5, last 5) design. To determine whether performance was different for Direct and Indirect feedback groups at transfer (day 2, 45°) we performed a mixed-model ANOVA with a 2 (group: Direct, Indirect) × 2 (trials: first 5, last 5) × 2 (training order: 30° then 60°, 60° then 30°) design. To examine whether performance was different for Direct and Indirect feedback groups during training, a further mixed-model ANOVA was performed with a 2 (group: Direct, Indirect) × 2 (trial time: first 5, last 5) × 2 [rotation size: small (30°), large (60°)] design.

Learning rate analysis.

It is possible that using an average of the first five and last five trials to examine learning could mask differences in the underlying learning curves of the groups. To better understand how MTs decreased across trials, we fitted a regression line to the data of each participant across their five trials to determine the rate of improvement at the beginning of each block (see Table 5). Another regression line was fit to the last five trials of each participant to see how much learning rate had reduced by the end (see Table 5). Because the patterns of findings were similar for MT and PL we did not repeat learning rate analysis on PL measures. To confirm that the last five trials at 0° had reached a comparable plateau we compared the gradients of the Direct and Indirect groups: there were no significant differences [t(42) = 1.04, P = 0.31], confirming that this was an appropriate baseline measure of MT.

RESULTS

At baseline the Direct feedback group (mean = 12.68 s, SD = 4.04 s) was significantly quicker [F(1,42) = 7.46, P < 0.01, η_p² = 0.15] than the Indirect feedback group (mean = 16.25 s, SD = 4.72 s). Not surprisingly, therefore, performance was better when participants could directly view their hand.¹ MTs did decrease from the first five to the last five trials [F(1,42) = 48.82, P < 0.001, η_p² = 0.54], reflecting a general improvement at performing the task, but there was no interaction with group [F(1,42) = 2.42, P = 0.13]. To study whether viewing the hand would help (through the provision of additional information) or hinder (via visual capture) motor learning, we studied performance across the first and last five transfer trials (where participants were exposed to 45° rotation on day 2). We found a significant main effect of group (Direct vs. Indirect) on 45° trials [F(1,40) = 35.78, P < 0.001, η_p² = 0.58], with the Indirect group showing improved performance (Fig. 2). There was also a main effect of first or last five trials [F(1,40) = 104.77, P < 0.001, η_p² = 0.72], demonstrating learning across trials. There were no significant differences between the large-small and small-large groups in either the Direct or Indirect conditions at transfer and no interactions (see Table 1 for full details of ANOVA). It seems that the rotation order at training had no consequences for performance on a novel rotation at day 2. The lack of interactions of group with first or last five trials suggests similar rates of learning for Direct and Indirect groups in the transfer session, but this is examined in more detail in Learning rate analysis. Our examination of PL (Table 2) confirmed the pattern of results observed in measures of MT, although the effect sizes observed were considerably smaller. This is presumably because participants were prioritizing movement accuracy over MTs so PL differences were generally small and on the order of 2 mm (average PL for Direct group was 25.4 mm compared with 23.4 mm for the Indirect group).

Fig. 2.

Average increase in time to complete trials (above baseline) for the Direct and Indirect groups for the first 5 trials (A and C) or last 5 trials (B and D) during training (day 1) or transfer (day 2) blocks. A and B show the data for the groups who performed the larger rotation (60°) before the smaller rotation (30°) during training, whereas C and D show the data for the groups who performed the smaller rotation (30°) before the larger rotation (60°) during training. On day 2 all individuals were tested on an intermediate rotation (45°) to see whether the training transferred to a novel visuomotor mapping. Bars = SE.

Table 1.

MT ANOVA results for day 2 transfer trials

	F	df	P	ηp2
Time (Tm)	104.77	1,40	<0.001†	0.72
(In)Direct (I/D)	35.78	1,40	<0.001†	0.47
Rotation order (RO)	2.80	1,40	0.10
Tm × RO	0.03	1,40	0.86
Tm × I/D	0.41	1,40	0.53
RO × I/D	0.07	1,40	0.79
Tm × RO × I/D	2.25	1,40	0.14

The groups (Indirect or Direct, small-large or large-small rotation order) were the between-subjects factors. MT, movement time. See Fig. 2.

^†Result significant at P < 0.001 level.

Table 2.

PL ANOVA results for day 2 transfer trials

	F	df	P	ηp2
Time (Tm)	12.86	1,40	0.001*	0.24
(In)Direct (I/D)	7.63	1,40	0.009*	0.16
Rotation order (RO)	1.72	1,40	0.20
Tm × RO	0.03	1,40	0.88
Tm × I/D	2.04	1,40	0.16
RO × I/D	4.43	1,40	0.042*	0.10
Tm × RO × I/D	0.76	1,40	0.39

The groups (Indirect or Direct, small-large or large-small rotation order) were the between-subjects factors. PL, path length.

^*Result significant at P < 0.05 level.

Formal analysis of training performance on day 1 revealed that all the main effects were significant, as were the interactions (Tables 3 and 4). ² Interpretation of main effects when there is a triple interaction [F(1,42) = 6.12, P < 0.05, η_p² = 0.13] should be carried out cautiously, but examination of Fig. 2 suggests that the Indirect groups produced shorter MTs than the Direct groups. It is possible that the observed differences between Indirect and Direct groups were simply driven by a single group, i.e., those who started day 1 with the 60° condition who performed particularly slowly. To ensure that this was not the case, we compared the final performance of just the Direct and Indirect groups at 30° (after having already completed 60°; Fig. 2A) and found that there were still significant differences in MT [t(20) = 5.04, P < 0.001].

Table 3.

MT ANOVA results for day 1 training trials

	F	df	P	ηp2
Time (Tm)	133.25	1,42	<0.001†	0.76
(In)Direct (I/D)	50.05	1,42	<0.001†	0.54
Rotation (Rot)	61.06	1,42	<0.001†	0.59
Tm × Rot	23.81	1,42	<0.001†	0.36
Tm × I/D	9.46	1,42	0.004*	0.18
Rot × I/D	7.56	1,42	0.009†	0.26
Tm × Rot × I/D	6.12	1,42	0.017*	0.13

The group (Indirect or Direct) was the only between-subjects factor. See Fig. 2.

^*Result significant at P < 0.05 level.

^†Result significant at P < 0.001 level.

Table 4.

PL ANOVA results for day 1 training trials

	F	df	P	ηp2
Time (Tm)	22.78	1,42	<0.001†	0.35
(In)Direct (I/D)	7.83	1,42	0.008*	0.16
Rotation (Rot)	14.91	1,42	<0.001†	0.26
Tm × Rot	9.45	1,42	0.004*	0.18
Tm × I/D	8.45	1,42	0.006*	0.17
Rot × I/D	5.81	1,42	0.020*	0.12
Tm × Rot × I/D	4.37	1,42	0.043*	0.09

The group (Indirect or Direct) was the only between-subjects factor.

^*Result significant at P < 0.05 level.

^†Result significant at P < 0.001 level.

Learning curves.

To understand group performance, the analyses so far have used average measures across the first five and last five trials. This has demonstrated that a rotated mapping impairs performance more when there is a direct view of the hand, both at the beginning and end of trials, and a direct view also leads to slower performance on a transfer task. Such performance measures are, however, coarse for investigating the rate of learning. The learning curves of each group are displayed in Fig. 3. It can be seen that the steepest learning seems to occur within the first five trials and that many groups have largely reached a plateau by the final five trials. To capture these changes, we fitted a linear regression to the first five and last five trials for each person to determine the rate of improvement in MTs. We then calculated the average gradient for each rotation and each group (Table 5).

To determine whether viewing the hand altered the rate of motor learning, we studied the gradients across the first and last five transfer trials (where participants were exposed to 45° rotation on day 2). We found no significant difference between the learning rates of groups [Direct vs. Indirect; F(1,40) = 0.003, P = 0.96] or rotation order [60°/30° vs. 30°/60°; F(1,40) = 0.061, P = 0.81]. There was a significant main effect of first or last five trials [F(1,40) = 43.37, P < 0.001, η_p² = 0.52], highlighting the reduction in learning rate across trials for all groups by the final five trials, but there were no interactions. We also wanted to see whether learning rates may have been different between groups on day 1. Because rates of learning can be elevated when performance is particularly bad (because there is greater opportunity to improve), we decided to compare just the Direct and Indirect groups who experienced the small rotation and then the large rotation (30°/60°; Fig. 3, C and D) and examine the first and last five trials for both rotations. The ANOVA showed that there were no Direct/Indirect group differences [F(1,20) = 0.097, P = 0.76]. There was a significant main effect between the first or last five trials [F(1,40) = 47.39, P < 0.001, η_p² = 0.70], demonstrating the reduction in learning rate across trials for all groups, but there were no interactions with group. There was a significant interaction between the size of rotation and the first or last five trials [F(1,20) = 5.81, P < 0.05, η_p² = 0.23], but post hoc tests revealed no significant differences between the rate of learning and rotation size over the first five trials [t(21) = 1.97, P = 0.06] or the last five trials [t(21) = 1.05, P = 0.31].

DISCUSSION

Visual information is invaluable for executing skillful manual tasks. At baseline (when no distortions were applied), the group who could see their hand while tracing (Direct group) were quicker than those who could not see their hand (Indirect group). It should also be noted that the visual display was in a different coordinate frame from hand motion in the Indirect condition, whereas in the Direct condition they overlapped. The baseline results suggest that there is an advantage to being able to see the controlled limb and have it move in the same coordinate frame as specified visually. Even though the individual signals provided from vision and haptics should have similar levels of noise, the combined estimate derived from congruent haptic and visual information will actually reduce the variability of information that is used to control skilled actions (Ernst and Banks 2002; van Beers et al. 2004). There were no differences in PL between Direct and Indirect groups and no change in PL over time, suggesting that both groups prioritized taking short straight-line trajectories, in some cases at the expense of slightly longer MTs.

When an additional distortion was applied to the mapping between vision and haptics (training and transfer), we observed a decrement in performance in all participants. Nevertheless, there was clear evidence that all of the participants adapted to the distorted input-output relationship over time, in line with previous reports (Abeele and Bock 2001). Critically, the data showed unambiguously that the Direct group experienced impaired adaptation relative to the Indirect group. This result may appear surprising because the Indirect group experienced a greater transformation between input and output,³ and it has been established that larger transformations are harder to learn (Abeele and Bock 2001). Because the Indirect group experienced the additional frame of reference transformation, it might have been predicted that this group would have experienced more difficulties with the task. Instead, it was the Direct group who had problems adapting to distorted visual-motor mappings when directly viewing the controlling limb, suggesting that visual capture was having a negative impact on performance.

These findings raise the question of the neurophysiological mechanisms by which visual capture is interfering with learning. The human nervous system is adept at incorporating multiple sources of information to control skilled actions and adjusting the weighting attached to information on the basis of its reliability (Ernst and Banks 2002; Mon-Williams et al. 1997a; Wilkie and Wann 2005). During the Direct viewing condition the combined visual-motor estimate will be much less noisy than during Indirect viewing, and so both visual and haptic signals are weighted highly (and so are difficult to ignore). In contrast, during the Indirect task the lack of visual signal related to hand position combined with the reference frame transformation (90° pitch of display) serves to add noise to the visual-motor signals, and so this may weaken the link between vision and haptics. According to this explanation, increased noise allows the nervous system to adjust the weight attached to haptic information and thereby speed the adaptation process (Burns and Blohm 2010; Burns et al. 2011; Sober and Sabes 2003, 2005). Because the Direct group experience less noise this might make the system less flexible in adapting the relationship between visual-motor input and output. The brain regions responsible for such adaptations have not yet been identified, though they are likely to involve both the parietal and premotor cortex (Graziano 1999).

An alternative interpretation of our findings is that the Direct group have to try to ignore irrelevant visual information to execute the task when a distortion is applied, whereas this information has already been filtered out for the Indirect group. We do not feel this provides a good framework for considering the findings since the crucial factor is not just the presence of irrelevant information but a particular property related to the visual presence of the hand. We have tested this informally by adding visual information to the Indirect display to represent the hand position, and this did not cause the same difficulties (though it provided a similar degree of visual distraction). It is also unclear at what point additional visual information should be treated as irrelevant. In principle, visual information about the hand could provide useful information about the degree of visual rotation applied to the display, and a priori it would be difficult to predict whether it should be classified as irrelevant. One other explanation for our findings is that removal of the end-effector improves performance because of a shift in “attentional focus” to the outcome of actions rather than the actions themselves. Such shifts in attention are very difficult to measure directly, but we think this explanation is unlikely because such an effect should have led to the Indirect group outperforming the Direct group even at baseline (whereas we actually observed the opposite effect).

These findings have obvious practical implications. Eye-hand coordination has been examined in laparoscopic surgeons: novices tended to fixate the laparoscopic tool, whereas the preferred strategy for experts was to fixate the target on the monitor (Law et al. 2004). There is a natural propensity for individuals to look at their hands when holding objects (Mon-Williams et al. 1997a), and this tendency might have the potential to make it more difficult for novices to learn how to control laparoscopic devices. Teaching surgeons to avoid fixating their hands when laparoscopic devices are being moved inside the patient's body might be a good place to start. Actively occluding vision of the hand could also be a useful intervention to prevent novice surgeons from fixating their hands. Our findings also suggest that engineering initiatives to generate “head up displays” (where the surgeon is provided with a transparent display that overlies the scene; Wagner et al. 1995) might be misguided. Laparoscopic surgery is just one example of a real-world task in which visual capture by the hand could lead to control problems. Technological advances mean that there are a multitude of robotic devices that require the user to learn a new motor mapping to use powerful machines (e.g., remote excavators). Some forms of teleoperative control effectively avoid the issue of disruptive visual capture by providing visual information through a head-mounted display where vision of the hand is unavailable (e.g., the “da Vinci” robotic surgical system). However, for cases in which vision of the hand is available it should be recognized that there may well be interference that will impair motor performance and learning. Moreover, merely obscuring the hand may not be sufficient to ensure maximally efficient movements. Our work shows that consistent vision and haptics (i.e., where there is no mismatch) result in the best performance, and effectively learning to operate a device without vision of the hand may require extensive training.

GRANTS

This work was partially funded by the Engineering and Physical Sciences Research Council (EPSRC) (UK) Bridging the Gap fund.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: R.M.W., R.L.J., and M.M.-W. conception and design of research; R.M.W., R.L.J., P.R.C., and M.M.-W. analyzed data; R.M.W., R.J.A., and M.M.-W. interpreted results of experiments; R.M.W. and P.R.C. prepared figures; R.M.W., R.J.A., and M.M.-W. drafted manuscript; R.M.W., P.R.C., R.J.A., and M.M.-W. edited and revised manuscript; R.M.W., R.L.J., P.R.C., R.J.A., and M.M.-W. approved final version of manuscript; R.L.J. performed experiments.