Is my hand connected to my body? The impact of body continuity and arm alignment on the virtual hand illusion
When a rubber hand is placed on a table top in a plausible position as if part of a person’s body, and is stroked synchronously with the person’s corresponding hidden real hand, an illusion of ownership over the rubber hand can occur (Botvinick and Cohen 1998). A similar result has been found with respect to a virtual hand portrayed in a virtual environment, a virtual hand illusion (Slater et al. 2008). The conditions under which these illusions occur have been the subject of considerable study. Here we exploited the flexibility of virtual reality to examine four contributory factors: visuo-tactile synchrony while stroking the virtual and the real arms, body continuity, alignment between the real and virtual arms, and the distance between them. We carried out three experiments on a total of 32 participants where these factors were varied. The results show that the subjective illusion of ownership over the virtual arm and the time to evoke this illusion are highly dependent on synchronous visuo-tactile stimulation and on connectivity of the virtual arm with the rest of the virtual body. The alignment between the real and virtual arms and the distance between these were less important. It was found that proprioceptive drift was not a sensitive measure of the illusion, but was only related to the distance between the real and virtual arms.
KeywordsVirtual hand illusion Rubber hand illusion Body perception Virtual reality Body representation Multisensory integration Virtual environments
In the rubber hand illusion (RHI) synchronous tactile stimulation of a person’s hidden real hand and a visible rubber hand in an anatomically plausible position results in an illusion of ownership over the rubber hand, a proprioceptive illusion of displacement of the real hand to the rubber hand, and a referral of the feeling of touch to the rubber hand (Armel and Ramachandran 2003; Botvinick and Cohen 1998).
It has been shown that even a computer-generated virtual arm in an immersive virtual environment (VE) can be integrated into the body representation if there is synchronous visuo-tactile correlation on the hidden real and the visible virtual arm that appears to extend from the real body of the participant (Slater et al. 2008). In this case not only was the perceptual system deceived by this virtual hand illusion (VHI), but also the motor system was recruited, inducing measurable muscle activity correlated with movements of the virtual arm. Furthermore, the VHI can be induced by visuo-motor correlations, even in the absence of visuo-tactile stimulation, where the movements of the tracked real hand are reproduced by movements of the virtual hand (Sanchez-Vives et al. 2010). In this case the visual feedback corresponds to the proprioceptive input and the motor output. The VHI has been also partially reproduced through a motor imagery based brain-computer interface, without the necessity for synchronized visuo-tactile or visuo-motor stimuli (Perez-Marcos et al. 2009).
The RHI, and by extension the VHI, depends on multiple factors. There is general agreement about the critical role of the visuo-tactile integration. If the visual and corresponding tactile sensations are not synchronous and well registered with one another in terms of location, the illusion does not occur (Botvinick and Cohen 1998; Slater et al. 2008). Another factor reported to be relevant for the RHI illusion is appearance, i.e. the resemblance between the subject’s real arm and the rubber one. The reported illusion is stronger for a rubber hand with a natural skin texture than one with a white latex glove (Haans et al. 2008).
Other important factors influencing the illusion are congruence, plausibility and connectivity. It has been shown that with the rubber arm placed in an incongruent position the strength of the illusion decreases (Ehrsson et al. 2004; Pavani et al. 2000; Tsakiris and Haggard 2005). The illusion of ownership is not only reduced with postural mismatches (10–30°) but also with stimulation mismatches between the real and virtual hand (Costantini and Haggard 2007). In the cases explored, not only was the rubber arm’s position incongruent with the real arm (from −10° to −180°) but also the arm was seen to be disconnected from the real body.
A less vivid illusion was also reported by those participants who were presented an ‘anatomically correct’ fake but distant hand, with the arm about twice as long as normal (Armel and Ramachandran 2003). In another study, the rubber arm was in a plausible position but incongruent with the position of the real arm (behind the participant’s back). This lessened the illusion as shown by a decrease in cross-modal extinction (Farne et al. 2000). The strength of the illusion has also been shown to decrease when the rubber arm is moved from its anatomically plausible position to the limits of reaching space (Lloyd 2007), while an impression of continuity with the body was given by covering up to the wrist with a hairdresser’s cape.
These findings and other not cited here (see “Discussion”) show a diversity of results regarding the relative position of the real and the fake arms and characteristics of the visuo-tactile correlations. Furthermore, when evaluating the responses to different experimental conditions, often body continuity is not taken into account, adding an implicit additional aspect to the experimental design.
Here we investigate for the first time the impact of four different factors on the strength of the VHI: (1) visuo-tactile synchrony (2) the degree of alignment between the real arm and the virtual arm (3) independently of alignment the amount of displacement between the real and virtual hands, and (4) body continuity in the sense of whether the virtual hand is connected or not to the rest of the body. Since earlier results have shown that the illusion of ownership over a virtual hand can be obtained in the same manner as over a rubber hand, we use virtual reality for this study, further illustrating its power and flexibility generally in the study of body ownership illusions.
Virtual reality system
Participants experienced a virtual body through a head-tracked stereo head-mounted display. This was a Fakespace Wide5, which has a field of view of 150° × 88° with an estimated 1,600 × 1,200 resolution displayed at 60 Hz. The head-tracking was realised with a 6-DOF Intersense IS-900 device. A 6-DOF Wand device was used by the experimenter to deliver tactile sensations to the real hand of the participant. The tracked Wand was represented in the virtual reality by a small yellow ball that was slaved to the movements of the real tracker. Hence when the real Wand touched the hand of the participant the virtual ball similarly was seen to touch the virtual hand in a corresponding location. The application was programmed using the XVR system (Tecchia et al. 2010) and the virtual body used the HALCA library (Gillies and Spanlang 2010).
There were three experiments and altogether 32 participants (22 females) were recruited. Their mean ± SD age was 24 ± 4. They had normal or normal-to-corrected vision and no history of neurological or psychological disorders. All but two subjects were right-handed. Handedness was assessed by the Edinburgh Handedness Inventory (Oldfield 1971). Upon arrival at the laboratory they were asked to read and sign a consent form, the experiment having been carried out in accordance with the regulations Comisión de Bioética de la Universitat de Barcelona. All the participants were paid 5€ for their participation in each session.
In each of the three experiments the experimenter tapped and stroked the real left hand of the participant with the Wand for 3 min while the virtual ball tapped and stroked the virtual left hand. Participants were asked to concentrate their attention on the back of the left virtual hand. In one condition of experiment 1 and all conditions of experiments 2 and 3 the seen virtual stimulation was synchronous with the real tactile stimulation, and registered at the same positions on the virtual hand as the stimulation was applied on the real hand. In experiment 1 there was an asynchronous condition, where a pre-recorded tapping and stroking motion of the ball on the virtual hand was used.
In each case the experiment was a repeated measures design with one factor at two levels (e.g., in the case of experiment 1 synchronous or asynchronous visuo-tactile stimulation). The conditions were applied in counter-balanced order across the participants. Details of each particular experiment are given in the “Results” section.
In experiments 1 and 2, after putting the HMD on and before the visuo-tactile stimulation was started, participants were instructed to close their eyes and point with their right index finger to the position of their left index fingertip. In order to avoid possible cues, the measurement was carried out over a box that the experimenter had placed over the participant’s left hand so that there was no physical contact between the fingers of the two hands during the measurement. Immediately after the 3 min of visuo-tactile stimulation participants were asked to repeat the action while still wearing the HMD and with closed eyes. In experiment 3, the pointing action was replaced by placing a piece of blue-tack under the table in the position corresponding to where they felt the centre of their palm to be, as described in Slater et al. (2008). In all cases, the horizontal distance between both markers, pre- and post-stimulation, corresponded to the proprioceptive drift. A positive drift meant a drift toward the virtual hand. Each proprioceptive drift measurement was based on two pointing actions—one pre- and one post-stimulation.
The time for an ownership illusion to occur, if at all, was recorded following Lloyd (2007). Participants had been instructed to say aloud when they felt for the first time, if ever, the virtual hand as their own. In any case the stimulation lasted for the full 3 min. Hence this ‘time to the illusion’ response is a censored variable, meaning that there was a cut-off of 180 s and that if the participant did not report the illusion by then, the experiment was stopped and their time to the illusion was recorded as 180 s. Hence in the results the mean time to the illusion is always an underestimate since it cannot be known at what point in time, if ever, those participants who did not report the illusion within 180 s might have reported it.
Q1. Sometimes I had the feeling that I was receiving the hits in the location of the virtual arm.
Q2. During the experiment there were moments in which it seemed as if what I was feeling was caused by the yellow ball that I was seeing on the screen.
Q3. During the experiment there were moments in which I felt as if the virtual arm was my own arm.
Q4. During the experiment there were moments in which I felt that if I moved my (real) arm the virtual arm would move.
Q5. During the experiment there were moments in which it seemed that my real arm was being displaced towards the right (towards the virtual arm).
Q6. During the experiment there were moments in which it seemed that the hits that I was feeling originated in some place in between my own arm and the virtual arm.
Q7. During the experiment there were moments in which I felt as if my real arm was becoming virtual.
Q8. During the experiment there were moments in which it seemed (visually) that the virtual arm was being displaced towards the left (towards my real arm).
Q9. During the experiment there were moments in which the virtual arm started to look like my own arm in some aspects (physically).
Q10. During the experiment there were moments in which I had the sensation of having more than one left arm.
Experiment 1: Synchronous/asynchronous
The first experiment Synchronous/asynchronous (S–AS) was the normal VHI but carried out using the HMD [the original VHI experiments used a stereo large single screen system with head-tracking (Slater et al. 2008)]. The hand was stroked with a ball attached to the hand-held wand that was tracked and represented in the virtual world as a yellow ball hitting the virtual hand (Fig. 1b). This was a single factor within-groups (repeated measures) design, the visuo-tactile stimulation being either synchronous or asynchronous, as described earlier. The virtual and real arms were parallel to one another (we refer to this as being aligned) with a distance of 20 cm between the positions of the real and virtual hands.
There were 15 participants in this repeated-measures experiment (mean ± SD age 23 ± 4; one male), 8 of whom received first the synchronous condition followed by the asynchronous (group SA) and the remainder in the opposite order (group AS). They were randomly assigned to these groups. Hence all participants completed both conditions in the order depending on their assigned group.
In each of the two trials, after the 3 min of tapping the subjects first carried out the proprioceptive drift task to test the illusion of displacement while wearing the HMD. Then they removed the HMD and completed the 10-item questionnaire (“Methods”).
In order to test for order effects (i.e., differences between the two groups AS and SA), a repeated-measures ANOVA was carried out on all of the response variables (Q1–Q10, the time to the illusion, and the drift). The results are shown in Supplementary Table 1. Since no order effects were detected, the results from both groups were combined for analysis (hence n = 30). Two-factor ANOVA was used, where one factor was the participant, and the other was the synchronous/asynchronous condition. Care was taken to ensure that the assumption of normality of residual errors was not violated, using the Jarque–Bera non-parametric test for normality (Jarque and Bera 1980), where a 0.05 significance level cut-off was used. The assumption of normality was rejected for three responses, Q5, Q10, and the drift. In each case two outliers were found by inspection of the residual errors, and when these were removed the Jarque–Bera test no longer rejected normality. In the case of Q5 only this resulted in the difference between synchronous and asynchronous being significantly different, but had no effect on Q10 or drift which both remained non-significant.
The time taken to report the onset of the illusion is significantly lower in the synchronous compared to the asynchronous condition (Fig. 2c). The proprioceptive drift was significantly greater than 0 over both conditions with mean value of 35.8 ± 42.7 mm (n = 30; mean ± SD; Fig. 2b).
Experiment 2: Connected/unconnected/not aligned
The purpose of this experiment was to test whether the VHI can be induced when the real and virtual arm are not aligned. The idea was that this would depend on whether or not the arm was connected to the rest of the body. Often in the rubber hand illusion the rubber hand is on the table, and clearly unconnected to the real body, although some experiments have used the device of partially covering the rubber arm and corresponding real shoulder with a cloth to give the impression of connectivity between the rubber hand and real body. Our hypothesis was that when the arm is visibly connected to the rest of the body the illusion of ownership would occur even when the real and virtual arms are not parallel to one another. In experiment 2 the left real arm was straight (Fig. 1a) and the virtual arm bent at 64° away from it (Fig. 1c). The response variables and design were otherwise the same as for S–AS (experiment 1). The estimated horizontal distance between the real and the virtual hand was 21.9 cm for males and 21.4 cm for females.
This was a repeated measures single factor experiment where the factor was connectivity of the virtual arm to the virtual body (unconnected or connected) (Fig. 1c, d). In both the connected and unconnected conditions synchronous stimulation only was used.
The number of participants with complete data was n = 13 (mean ± SD age 24 ± 5, one male), a subset of those who took part in experiment S–AS. Experiment 2 was carried out more than 30 days after experiment 1. Six participants experienced first the connected and then the unconnected conditions (group CU), and the remainder in the opposite order (UC). Hence all participants completed both conditions in the order given by the group to which they were assigned. The order effects were explored using ANOVA (Supplementary Table 2). Since no order effects were detected, the results from both groups were further analysed together (n = 26).
There were no differences in the proprioceptive drift between the connected and unconnected conditions, the mean drift being 50.5 ± 48.8 mm (n = 26; mean ± SD) (Fig. 3b). However, the drift over both conditions was significantly greater than 0. Finally, the mean time to experience the illusion was significantly less in the connected than in the unconnected condition (Fig. 3c).
This experiment suggests that the illusion of ownership can be evoked by the synchronous visuo-tactile stimulation of a real and a virtual arm, even when they are not aligned (the real arm straight and the virtual arm bent), and that this is aided by the virtual arm being connected to the body (Fig. 1c, d).
Experiment 3: Connected/unconnected/co-located
This experiment also investigated the difference between connected (Fig. 1c) and unconnected (Fig. 1d) conditions. The crucial difference was that in this experiment the virtual and real arms were coincident, and therefore aligned and with no displacement at all between real and virtual hand positions. In this case the real arm was also bent 64°, co-located with the virtual arm as in Fig. 1c. This was carried out with a different set of participants.
Seventeen participants (mean ± SD age 25 ± 4; 9 males) were recruited, with 8 experiencing first the connected and then the unconnected connected conditions, and the remainder in the opposite order. Otherwise everything was the same as the previous experiment, including the fact that all stimulations were synchronous. Hence all participants completed both conditions in the order given by the group to which they were assigned. Supplementary Table 3 shows the within-groups ANOVA for experiment 3. There were no order or interaction effects and therefore the results were analysed together (n = 34).
The time to perceive the illusion was also significantly less for the connected than for the unconnected hand (Fig. 4c). As expected, since both the virtual and real arms were co-located, the proprioceptive drift was not significantly different from 0 in any condition (Fig. 4b).
Comparisons across all conditions
In the experiments described above we have considered individually the impact of various factors on ownership: visuo-tactile synchronous stimulation, alignment, and body connectivity. Implicitly there was a fourth factor, which was the displacement, i.e., the distance between the real and virtual hands in each case. What is of most interest is not these factors considered separately but their relative effects when considered together. It is possible to explore this by combining all the data together and carrying out a regression analysis of the response variables (Q1–Q10, time and drift) on all four factors (n = 90): (1) Synchrony versus asynchrony of the visuo-tactile stimulation, (2) Alignment of the virtual with respect of the real arm, regarding whether or not the arm was parallel or not with the virtual arm. (3) Displacement, i.e. the distance between the position of the real and virtual hands, and (4) Body continuity (connected vs. unconnected).
Experiment 1: displacement = 20 cm, alignment = 1 (parallel); Synchronous versus Asynchronous; Connected; n = 30
Experiment 2: displacement > 21 cm, alignment = 0 (at an angle); Synchronous; Connected versus Unconnected; n = 26
Experiment 3: displacement = 0 cm, alignment = 1 (coincident); Synchronous; Connected versus Unconnected; n = 34.
Regression analyses on combined data using robust regression
The time that it took for the illusion to occur was influenced by two factors. This time was reduced by synchronous stimulation and by connectivity. Other things being equal synchronous compared to asynchronous on the average reduced the time to the illusion by 89 s. Similarly, connectivity reduced the time by about 47 s.
The proprioceptive drift was only associated with displacement, such that the greater the distance of the virtual to the real hand, the greater the drift. This is to be expected and should be considered as a consistency result.
Our analysis also suggests that alignment does not contribute to the illusion of ownership, but does contribute to referral of touch (Q1).
In this study we have demonstrated the influence of body continuity and arm alignment on the illusion of ownership over a virtual hand. By body continuity we mean the visual connection between the body and arm. We find that body continuity can compensate for other factors such as misalignment. A mismatched or misaligned (64°) virtual arm with respect to the real arm while in a plausible position can evoke an illusion of ownership when the body continuity is preserved.
It is found that the time to evoke the illusion is sensitive to the different experimental conditions, while the proprioceptive drift is less sensitive given that it is a function only of the displacement between the real and the virtual arms. Lloyd and colleagues earlier reported a strong nonlinear relationship between the time taken to rate the strength of the illusion and the position of the rubber hand (Lloyd 2007), this time being shorter for closer hand positions.
The illusion of positional displacement or proprioceptive drift has been classically considered an objective measurement of the RHI since Botvinick and Cohen (1998). The VHI also induces a proprioceptive drift towards the virtual arm (Slater et al. 2008). However, the relationship between the subjective illusion of ownership and the proprioceptive drift is a matter of debate. Holmes et al. (2006) found that the subjective illusion of ownership of a rubber hand did not correlate with the proprioceptive recalibration, which existed in cases when the illusion was not reported. Haans et al. (2008) also found proprioceptive displacement following visuo-tactile synchronous stimulation of a table top and the real hand, even when no RHI for the table top was evoked. A recent paper by Rohde et al. (2011) further demonstrates that there may be a proprioceptive displacement while there is no illusion of ownership. Proprioceptive displacement was evoked not only by just the vision of the rubber hand—and no visuo-tactile correlated stimulation—but also during short periods of asynchronous stimulation. In their experiments, only long (120 s) asynchronous stroking had the power to block proprioceptive displacement. Further evidence that visuo-tactile stimulation is not necessary for proprioceptive displacement is provided by Durgin et al. (2007), where stroking of the fake hand with the light from a laser pointer, even in absence of real touch induced ownership and proprioceptive displacement.
The results of our experiment suggest that proprioceptive displacement may occur even when there is no subjective illusion of ownership. This was the case in the asynchronous visuo-tactile condition. Even when proprioceptive displacement has been found for short asynchronous stimulation (Rohde et al. 2011) to our knowledge this is the first time that significant proprioceptive displacement has been reported for continuous asynchronous stimulation. How can we explain this? We suggest that this is due to the relevance of first person perspective provided by a HMD. In the HMD the visual information is that of a body which is co-located with one’s own and of two arms that are connected to the virtual body. Our previous studies using a HMD found that the first person perspective is the most important factor in order to feel a whole virtual body as one’s own (Slater et al. 2010b). It would seem that the first person view of one’s own body is powerful enough to counteract the visuo-tactile conflict during asynchronous stimulation. It should be mentioned though that even calibrated HMDs are not free of geometric distortions (Kuhl et al. 2009) which could have some influence on the estimation of the relative position of the virtual with respect to the real hand. However, it has also been found that such egocentric distance judgments improve when the participant is represented by an avatar (Mohler et al. 2010).
The control questions
In addition to the illusion related questions Q1–Q3, we found that three of the control questions had high scores—Q5 (the feeling of displacement of the real arm towards the virtual), Q7 (the feeling of the real arm becoming virtual) and Q9 (the feeling of the virtual arm coming to look like the real arm). This reflects what was also found in Slater et al. (2008). Considering some of the questions as ‘controls’, in the sense of being unrelated to the illusion of ownership, is likely to be unsupportable. In fact there is no reason to believe that the illusion should not be accompanied by the types of sensation reported in those questions, and researchers in the field continue to use these questions largely for the historical reason that they were used in the original RHI paper (Botvinick and Cohen 1998).
Alignment and body continuity
While alignment between the real and rubber arm has been explored in the literature, this has not been the case with body continuity. Body continuity or connectivity refers to the existence of an apparent connection between the body and the virtual arm and hand. A common strategy to achieve this with a rubber arm has been to create an illusion of body continuity by a cloth that gives the impression of a sleeve and provides an illusion of continuity between the body and the rubber arm. However, the use of a HMD obviates the need for this, since the whole body as a connected structure seen from a first person position can be attained. One possibility is that the fact that the whole environment is virtual, the body is part of this environment, and this enhances the probability of an ownership illusion. For example, results in Slater et al. (2010a) suggested that having a virtual body in such an immersive virtual reality enhances presence, the sense of being in the virtual space, and also the illusion that what is happening there is real.
It could be argued that since participants have to concentrate their gaze and attention towards the fake hand that this fact could avoid the need for a full virtual body, since it may not even be noticed during the stimulation—and this would definitely be the case if a narrow field-of-view HMD were used. However, in the case of our experiment, a wide field-of-view HMD was used so that the virtual body was seen in peripheral vision. Also, at various times participants would clearly have seen the connectivity or lack of it, for example during the initial period of looking around the scene. Indeed, our results show that not only there is a larger subjective ownership illusion with body connectivity, but also the time to evoke the illusion was significantly reduced.
There are differing results in the literature concerning the importance of the spatial relationship between the real and fake arm with respect to inducing the illusion of ownership. For example Ehrsson et al. (2004) used as a control condition not only asynchronous stroking but additionally the rotation of the rubber hand by 180° so that it was pointing towards the body of the subject, arguing that the RHI occurs for synchronous stroking but also “when the rubber hand is aligned with the subject’s own hand”. This followed the finding reported by Pavani et al. (2000) which found that the illusion required the rubber hand to be spatially aligned with the real hand, where spatial alignment in this case referred to a rotation of the position of the rubber hand with respect to the real hand. In both the Ehrsson et al. and Pavani et al. the rubber hand was not connected to the body. In the case of Ehrsson et al. the hand was additionally not in an anatomically plausible position (pointing towards the subject) and arguably in Pavani et al. it was.
Tsakiris and Haggard (2005) also found that flexion and displacement of the rubber hand with respect to the real hand eliminated the illusion, while Costantini and Haggard (2007) found that the illusion remained if there were small displacements of the real hand with respect to the rubber hand, but not when these were applied to the rubber hand. Again in both of these cases it appears that there was not an attempt to make an apparent connection between the rubber hand and the body of the subject.
Lloyd (2007) described an experiment where the rubber arm was placed in one of 6 different anatomically plausible positions (within peripersonal space). In this case the rubber arm was both rotated and translated away from the real hand. It was found that the illusion diminished with distance, based on both a subjective questionnaire measure, and time to the illusion. The rubber arm was covered by a cloth so that it did appear to connect to the rest of the body. However, Zopf et al. (2010) found that the illusion was not diminished when there was a lateral translation of the rubber hand away from the real hand, in this case with no rotational component. Again the rubber hand was in a plausible position, and was covered by a cloth to make it seem connected to the body.
In our experiments we have found that all three illusion related questions Q1–Q3 are dependent on synchronous visuo-tactile stimulation. However, Q1 (referral of touch) was also dependent on alignment (i.e., the real and virtual arms being parallel). Q3 (ownership) was dependent on body connectivity, as was time to the illusion. All of our conditions involved the virtual hand in an anatomically plausible position. Our conclusion is that for the illusion of ownership of the virtual hand body connectivity should be considered as an additional important variable. We speculate following Slater et al. (2010b) that seeing the virtual body from a first person perspective is also critical, although this variable was not manipulated in the current experiment. We also conclude from these experiments that proprioceptive drift should be downgraded as a measure of the illusion. When there is first person experience of a whole connected body in immersive virtual reality then this becomes a dominating factor with respect to proprioception and that drift simply records the degree of separation between the real and virtual arms. Perhaps there is a tolerance region where not too much separation results in the virtual and real arm being sensed in the same place with respect to proprioception.
We thank Bernhard Spanlang for providing the avatar library (HALCA) and Jean-Marie Normand for his help during the experiments. Konstantina Kilteni provided resources that helped in the literature review. This research was supported by FP7 EU collaborative project BEAMING (248620) and the ERC project TRAVERSE (227985).
- Slater M, Perez-Marcos D, Ehrsson HH, Sanchez-Vives M (2008) Towards a digital body: the virtual arm illusion. Front Hum Neurosci 2. doi:10.3389/neuro.3309.3006.2008
- Slater M, Spanlang B, Corominas D (2010a) Simulating virtual environments within virtual environments as the basis for a psychophysics of presence. ACM Transactions on Graphics (TOG) - Proceedings of ACM SIGGRAPH 2010. Paper 92Google Scholar
- Slater M, Spanlang B, Sanchez-Vives M, Blanke O (2010b) First person experience of body transfer in virtual reality. PLos ONE e10564. doi:10510.11371/journal.pone.0010564