ArmDeformation: Inducing the Sensation of Arm Deformation in Virtual Reality Using Skin-Stretching

2.1 Body Ownership

Body ownership is a well-studied topic in the fields of psychology  [16] and neuroscience  [69, 85]. Previous research has shown that the sense of body ownership arises from integrating multisensory information, including visual, haptic, and proprioceptive cues  [27, 45].

In Human–Computer Interaction (HCI) research, body ownership has been studied in VR and Augmented Reality (AR) applications  [13, 30]. One commonly used technique for inducing body ownership illusions is the Rubber Hand Illusion  [15, 78], in which participants see their hidden hand being stroked while simultaneously viewing a rubber hand being stroked. This creates the illusion that the rubber hand is part of their own body. Sungchul Jung et al.  [42] discovered that using customized virtual hands improved participants’ ability to estimate the size of virtual objects and increased participants’ sense of virtual body ownership and spatial presence compared to using generic virtual hands. Irene Senna et al.  [74] gradually replaced the sound of a small hammer striking skin with the sound of a hammer striking a block of marble while continually gently striking participants’ hands with the hammer to create a false perception of the material characteristics of the hand.

Research on haptic feedback systems has also explored ways to enhance the sense of body ownership in VR. Valentin Schwind et al.  [73] conducted a VR user study to investigate how the brain combines visual and haptic signals in response to different hand appearances using a cue-conflict paradigm. According to Andrea Webb Luangrath et al.  [56], the active aspect of product contact, which results in a notion of physical ownership of the virtual hand, influences consumer psychological ownership and product valuation. The "slime hand illusion"  [49] is a method where participants watch an experimenter manipulate slime in a mirror while having their hidden hand handled similarly. This causes a nonproprioceptive ownership distortion and gives the participants the impression that their finger or hand is being stretched or deformed like the visible slime.

In summary, research has shown that the illusion of body ownership can be induced in reality and VR through coordinated sensory feedback. Yet, there needs to be more literature concerning the specific illusion of forearm bending and its influence on users’ perception of body ownership in VR. In our paper, we align with the existing research that explores body ownership during the body deformation illusion, and we also explore how varying forearm bending degrees impact the sense of body ownership.

2.2 Body Illusion

To enhance users’ immersion with deformed avatars in VR and reality, previous research gives users visual feedback or haptic feedback that allows them to better adapt to the avatar  [13, 45, 46, 87, 93].

Previous perceptual psychology research suggests that when proprioception and vestibular sensation are not aligned, vision tends to dominate them  [14, 38]. Therefore, visual feedback can make the user adapt to the avatar. Gulliver’s virtual travels  [75] examine the effects of body size manipulation on user experience. The results show that reducing the body avatar size resulted in a noticeable decrease in participants’ body image perception, while no significant change was observed when zooming in on the body. Being Barbie  [87] showed that participants who experienced owning a Barbie-sized body perceived objects as larger and farther away, while those who experienced owning a larger body perceived objects as smaller and closer. Mar Gonzalez-Franco et al.’s  [30] experiments show that users unconsciously follow their avatars as long as their virtual body does not overlap with their physical body. Beyond Human  [51] proved that taking on nonhuman morphologies (including extra body parts and access to their respective superhuman skills) would lead to high levels of gameplay enjoyment. Andrea Stevenson Won et al.  [92] demonstrated that people can quickly learn to use a new avatar and succeed in that avatar, even if the avatar model is completely different from the user’s own body.

Although the user can adapt to the mismatch between the avatar body and the physical body, providing some haptic feedback can enhance the immersion of the user avatar. Christopher C. Berger et al.  [13] recreated a physical illusion in VR through which the perceived length of a person’s nose and arm was extended. Participants tapped themselves on the tip of the avatar’s nose, as seen from a first-person perspective, and the avatar’s nose slowly grew longer with each tap. Konstantina Kilteni et al.  [46] explored the extent of user ownership over a virtual arm when extended beyond the length of their real arm. The findings indicate that participants felt a sense of ownership over the virtual arm when it was up to three times longer than their real arm, but this feeling weakened when the virtual arm was elongated up to four times the length of their real arm. Majed Samad et al.  [72] reproduced the rubber hand illusion and demonstrated that the experiment could be achieved without the sense of touch, but that synchronized stroking would enhance it.

Unlike earlier research that focused solely on the elongation of specific body parts  [13, 46, 93] or the overall enlargement or reduction of the entire body  [75, 87]. Alternatively, only practical applications of forearm bending are considered  [7], and research on body ownership after forearm bending is missing. In contrast, our work explores the effect on body ownership when the forearm is bent. While earlier research favored the psychological study of body deformation illusion, our work combined this with practical applications, and we presented several possible application scenarios.

2.3 Skin-Stretching Feedback

Currently, haptic feedback is commonly used in both reality and VR, and much research has used haptic feedback to enhance user immersion  [48, 77] or provide additional guidance information[35, 84]. And the following are some common ways to enhance haptic feedback: skin-stretching  [12, 22, 36, 59, 65, 89], vibration  [11, 25, 43], force  [18, 21, 91] and so on. Our paper mainly uses skin-stretching to create the illusion of arm deformation.

There are two main applications of haptic feedback from skin-stretching: one is to enhance the user’s sense of reality  [77, 89, 93], and the other is to convey information  [17, 19, 20]. An example of skin-stretching is Gum-gum shooting  [93], which uses a stretching simulation device attached to the forearm. The operation of the device depends on the rotation of two motors to lengthen and restore the skin of the forearm, which can make the user feel that the forearm is longer. Masque  [88] applied skin-stretching technology to the face. It stretches the skin of the face to simulate the bumps and inertia of riding a motorcycle. Skin Stretch Stylus  [68] affects the user’s perception of surface stiffness by stretching the skin. When the device is pulled down, the intensity of the skin-stretching will be according to the normal force applied to the virtual surface, which simulates different degrees of stiffness of the virtual object the device is in contact with.

Nathaniel A. Caswell et al.  [17] developed a device that can apply the directional cues to the forearm and doesn’t need to occupy the fingers. Francesco Chinello et al.  [20] designed a lightweight hand strap with four actuators, each driving a cylindrical actuator that can rotate to perform skin-stretching and use the information from the skin-stretching as navigation. In addition to directional information, skin-stretching can convey geometric shapes or characters. Skin Drag Displays  [39] can produce two different stimulus types instead of the vibrotactile array, which enables the user to recognize tactile shapes through skin-stretching better. tactoRing  [41] is a new tactile display that allows for more accurate cue recognition by dragging a small tactile device over the skin around the finger and stimulating multiple skin areas.

Unlike previous work that only stretched the skin in a fixed area of the forearm in a specific direction, we innovatively designed a device that allows the skin on any side of the forearm to undergo skin-stretching in either of two directions. Concurrently, we integrate this skin-stretching with the illusion of arm deformation. Our approach differs from research like Gum-gum shooting  [93], as we conducted perception studies before designing our wearable device. This exploration helped us determine the optimal number of skin-stretching actuators and the designs of skin-stretching that could most effectively induce the illusion of arm deformation.

3.1 ArmDeformationBox

This study used a unique ArmDeformationBox firmly fixed to a table. The device is a layered structure comprising three skin-stretching layers. The layers are arranged at intervals of 60 mm. Each layer adopts a square shape (290mm × 290mm) with a 150 mm diameter circular hole at the center, allowing the forearm to pass through. Two actuators (MG996R) are positioned on each skin-stretching layer’s upper and lower sections. These actuators can be adjusted along the track to fit the thickness of each user’s forearm. Each actuator propels a silicone pad with a 30mm diameter, which can stretch the user’s skin up to a maximum distance of 20mm (10mm inward and 10mm outward)  [88]. To ensure forward-and-backward movement of the silicone pad for skin-stretching feedback in our design, we designed the gap between each layer to be 60 mm. We also took into account the user’s forearm length, which ranges from 252 to 320mm for males and 225 to 298mm for females  [34, 94]. Based on these factors, we determined that the maximum number of layers should be three, consisting of six actuators. The elastic cord connecting the two actuators can be easily adjusted to vary the pressure on the user’s forearms, ensuring user comfort and effective skin-stretching.

Figure 2:

View Figure

3.2 Procedure

Figure 3:

View Figure

We begin by having the user pass their dominant hand through the ArmDeformationBox and hold it steady. Visual distractions emanating from the ArmDeformationBox were suppressed using an HTC VIVE Pro Eye head-mounted display (HMD) to optimize focus on the experimental stimuli. The experimenter then made the actuator tightly in contact with the skin using the elastic cords. In VR, a virtual version of their forearm appeared. In this experiment, participants are made to experience an illusion of their forearm either elongating or shortening, as shown in Fig. 3. Crucially, they were instructed to maintain a consistent forearm posture during skin-stretching. The sensory feedback was varied: participants received either a visual with haptic feedback or visual-only feedback. The haptic feedback also varied, with one, two, or three actuators activated per side. Notably, the experimental design was within subjects, meaning every participant experienced five iterations across these four conditions. The silicone pad is adjusted from the center to - 10mm (inward) or 10mm (outward), which is mapped visually 1/2 times (shortening) or two times (elongating) the original forearm length, respectively.

The experiment subjected participants to four conditions presented randomly. Each condition differs in the number of actuators used to render skin-stretching along the hand:

All haptic conditions start at the home position and gradually move to maximum skin-stretching distance within an equivalent time frame. At the same time, the visual forearm’s deformation mirrored the intensity of the skin-stretching. After the participants answered the question, the silicone pad returned to its home position.

The directionality of the actuators’ movement was aligned with the illusion. In simulations of forearm elongation, the actuators pushed the silicone pads outward. Conversely, in the illusion of forearm shortening, the movement was inward.

A post-stretch assessment phase followed each trial, during which participants needed to respond to a series of questions. To probe the effects of the body deformation illusion on the user experience in VR, we employed the embodiment questionnaire created by Peck et al.  [63] (see 3.4). This questionnaire consists of seven Likert scales, where responses span from -3 (strong disagreement) to 3 (strong agreement). We selected this questionnaire due to its user-friendliness and adaptability. Prior research has evidenced that subjective embodiment measures, garnered through questionnaires, align with objective metrics, such as electroencephalograms  [32]. Embodiment intersects with other key factors like presence  [61], perception  [8, 29], and ultimately behavior  [28]. Therefore, it can be a useful metric that also helps discuss the effects of the body deformation illusion on other critical aspects of the VR experience. Adopting a comprehensive embodiment questionnaire was our first step in revealing the effects of the body deformation illusion on body embodiment in VR  [31, 63, 76, 83].

Participants used a VIVE controller to answer seven questions by moving a slider bar to select a score between -3 and 3, as shown in Fig.  3. Participants were engaged in this structured experiment for roughly 40 minutes and allowed to take a 5-minute break at any point to prevent potential fatigue.

3.3 Participants

In the forearm elongating illusion experiment, we recruited 18 participants (8 self-identified females, 10 self-identified males) aged 20-27 (M = 24.2, SD = 2.2), with forearm length (including the hand) ranging from 400 to 450 mm (M = 427.2, SD = 18.7), all right-handed. Among them, 13 had previous VR experience, with 3 familiar with visuo-haptic illusions. For the forearm shortening experiment, we recruited 18 participants (8 self-identified females, 10 self-identified males) aged 22-27 (M = 24.9, SD = 1.8), with forearm length (including the hand) spanning 400 to 480 mm (M = 431.7, SD = 20.3) and all of them are right-handed. Of these participants, 15 had prior VR experience, and 4 were acquainted with visuo-haptic illusions. All participants were compensated with a locally equivalent amount of 10 USD for their involvement. The study was approved by our Institutional Review Board.

3.4 Results and Findings

Figure 4:

View Figure

To probe the effects of the body deformation illusion on the user experience in VR, we employed the embodiment questionnaire created by Peck et al.  [63], consisting of seven Likert scales ranging from -3 (strong disagreement) to 3 (strong agreement):

As underscored in prior research, the measured scores typify participants undergoing a profound embodiment sensation [31, 63, 83]. The degree to which participants felt the virtual body was their own, known as embodiment score, is calculated as: \(\begin{align*} &\text{Embodiment Score}\\ & = \cfrac{((Q_1-Q_2)/2*2+Q_4*2+(Q_3-Q_5)/2*2+Q_6+Q_7)}{8} \end{align*}\)

Where the weights of the categories are based on their perceived importance in the overall embodiment experience [44, 64].

By calculating the above equation, we can get the following results. In the case of the forearm elongating, visual-only (M = -0.715, SD = 0.702), two actuators (M = 0.675, SD = 0.771), four actuators (M = 0.974, SD = 0.868), and six actuators (M = 0.997, SD = 0.827); in the case of forearm shortening, visual-only (M = -0.499, SD = 0.990), two actuators (M = 0.521, SD = 0.670), four actuators (M = 0.613, SD = 0.696), and six actuators (M = 0.535, SD = 0.665).

We used the Friedman Test post-Wilcoxon signed rank test (See Fig. 4) which shows significant differences between visual-only and visual with haptics (p < 0.05); however, there were no significant differences between the haptic groups (p > 0.05) except for one group (two actuators vs six actuators with the forearm elongating illusion(p = 0.011)).

First, the embodiment sensation is statistically significantly higher in all cases with skin-stretching feedback compared to visual stimuli only. These results confirm that skin-stretching feedback can significantly improve body ownership when forearm deformation is applied in VR. This aligns with previous research showing that body ownership improves when provided with haptic feedback  [52, 72]. Second, in most cases, there was no significant difference between the cases with haptic feedback. So, we decided to use two actuators for our final wearable prototype. To make the decision, we also consider the cost-effectiveness and weight of the device. We considered that the fewer actuators we use, the lighter and more comfortable the device will be. This decision was made keeping in mind the effectiveness of the illusion, the practicality of the device, and user comfort.

4.1 Mathematical Model for Visual Forearm Bending

We define a bend of the forearm by looking at the angle between the arm axis near the elbow and the axis direction near the wrist (See Fig.  5). The new curved axis is a circle arc tangent to the two directions at both ends of the bent forearm. The model of the forearm, as it is rendered in the Unity game engine we use, is deformed by calculating a new position for each mesh vertex. The calculation of the geometric deform is described in appendix  A.

4.2 Procedure

Figure 5:

View Figure

Based on Perception Study 1, we concluded that using two actuators can effectively generate skin-stretching. So, we set our hardware to render the skin-stretching on the top and bottom of the forearm. This study aimed to find the optimal skin-stretching design for forearm bending.

For this study, we utilized the same ArmDeformationBox as in Perception Study 1 with a single skin-stretching layer: two actuators on the upper and lower sides of the forearm. Another difference from Perception Study 1 is that as we render forearm bending, the actuators will render opposing directions of skin-stretching.

In a pilot study with four participants (3 self-identified males and 1 self-identified female, 22-25 years old, SD: 1.25) With the same procedure in study 3 inspired by  [4, 58], we determined that participants maintain embodiment while their forearms are rendered bending between an upward bending of 56.3° and a downward bending of 45.2°. So, we decided the bending range was an upward bending of 45° and a downward bending of 45°. In Perception Study 3, we conducted a detailed visual embodied bending threshold study with the skin-stretching design that we found in this study.

Similar to Perception Study 1, We begin by having the user pass their dominant hand through the ArmDeformationBox and hold it steady. The participant’s visual display shows their forearm gradually bending, while ArmDeformationBox actuators gradually stretch the skin on top and bottom of the forearm. For the case of the forearm bending upward, the top actuator drives the silicone pad inward, and the bottom actuator drives the silicone pad outward to simulate the changes in skin lengths due to the bending. When the forearm is bent downward, the actuator flips accordingly.

The participants experience bending at randomized directions, and the haptic rendering is randomly chosen out of 4 conditions: full rendering where both top and bottom actuators are rendering corresponding and opposing skin-stretching forces, rendering using the top actuator alone, using the lower actuator alone, and a control condition where neither actuator is moving. While the order of conditions is random, each condition is repeated five times.

Each forearm bending condition requires reaching the maximum angle of 45° from 0° in the same time frame, synchronized with a gradual increase of the skin-stretching to a maximum of 10mm each time. After reaching maximum visual forearm bending and maximum skin-stretching distance, participants answered the same seven body ownership questions as in Perception Study 1. Participants used a VIVE controller to answer seven questions by moving a slider bar to select a score between -3 and 3, as shown in Fig.  3. After participants answered the questions, the actuators returned to the home position.

4.3 Participants

We recruited 18 participants (8 self-identified females, 10 self-identified males) aged 21-27 years (M = 23.8, SD = 1.7), all right-handed. Their forearm length (including the hand) varied between 410 and 480 mm (M = 429.4, SD = 21.5). Of these, 13 participants reported prior experience with VR technology, and six were acquainted with visuo-haptic illusions. All participants were reimbursed with an amount equivalent to 10 USD in their local currency for their participation. Our Institutional Review Board approved the study.

4.4 Results and Findings

Figure 6:

View Figure

We used the Friedman Test post-Wilcoxon signed rank test, as shown in Fig.  6. When the forearm bending upward, the visual-only condition scored (M = -1.067, SD = 0.998), both actuators’ condition scored (M = 0.558, SD = 0.823), the top actuator-only condition scored (M = -0.072, SD = 0.703), and the bottom actuator-only condition scored (M = 0.063, SD = 0.649). When the forearm bending downward, the visual-only condition scored (M = -1.141, SD = 0.999), both actuators’ condition scored (M = 0.780, SD = 0.758), the top actuator-only condition scored (M = 0.513, SD = 0.794), and the bottom actuator-only condition scored (M = 0.203, SD = 0.683).

Our results indicate that using both actuators together performs better than using only one actuator or relying on visual feedback alone (p < 0.05) in both upward and downward bending scenarios (as shown in Fig.  6). This is likely because two actuators can provide more complete haptic feedback, which helps the user feel like their forearm is bending. Consequently, we chose to use two actuators operating synchronously to evoke the forearm bending illusion.

5.1 Procedure

We utilized the ArmDeformationBox as in the Study 2. In this study, we implemented a method involving the synchronous movement of two actuators to simulate the forearm bending deformation. During the experiment, participants experience an illusion of forearm bending at different angles, ranging from 20° to 160° in increments of 20° , with each condition repeated four times. The experiment was categorized into upward and downward bending segments. In each trial, the actuators synchronously drove the silicone pad from its home position to the maximum distance within the same time frame, providing consistent haptic feedback stimuli when bent in the same direction.

Earlier research  [4] and we are both focused on illusions in VR, aiming to pinpoint the threshold above which users strongly believe the illusion. However, our unique interest lies in understanding when users feel most connected or possessive of the virtual forearm. To dive deeper into this aspect, we tweaked the original questions from previous research  [4, 58] and asked participants to answer the two following questions after each trial:

Participants used a VIVE controller to respond to these two questions as Fig.  3 shown. For the first question, participants answered either "yes" or "no," while for the second question, participants moved a slider bar to select a score from 1 to 5.

5.2 Participants

We recruited 16 participants (6 self-identified females, 10 self-identified males) aged 20-27 years (M = 23.9, SD = 2.2), all right-handed. Their forearm length (including the hand) varied between 400 and 480 mm (M = 438.1, SD = 27.9). Of these, 13 participants reported prior experience with VR technology, and six were acquainted with visuo-haptic illusions. All participants were reimbursed with an amount equivalent to 10 USD in their local currency for their participation. Our Institutional Review Board approved the study.

5.3 Results and Findings

Figure 7:

View Figure

Referred from previous work [4, 58], to calculate the bending threshold for every participant, we divided the instances wherein the participant experienced the virtual forearm as their own at that specific angle by the total repetitions as the Body Ownership Rate. The concept of Body Ownership Rate refers to the extent to which a user perceives a virtual forearm as a natural extension of their own body. A participant was deemed to have experienced the illusion of the virtual forearm being their own when they chose "yes" for the first answer and reported a confidence level of 3 or higher. Following this, we averaged the confidence rates across participants for each angle and charted the results. We then employed a threshold estimation curve of the form (where a and b are constants): (1) \(\begin{equation} f(x) = \cfrac{1}{1+e^{ax+b}} \end{equation}\)

As previous research  [4, 58, 81] suggests 0.75 is a reasonable acceptance rate, we chose the angle value corresponding to an average Body Ownership Rate equal to 0.75 as the threshold of forearm bending angle.

For upward forearm bending, with constants a = 0.0313, b = −2.6490, the threshold was determined to be 49.4830°. For downward forearm bending, with constants a = 0.0361, b = −2.9244, the threshold was calculated as 50.5061° , as Fig.  7 shows.

These findings suggest that the range of visual forearm bending should be considered in developing forearm bending VR applications with haptic feedback. We designed the applications we propose hereafter by comprehensively considering our study results and the results of previous research  [46] on arm-length adjustment.

6.1 Hardware

The device’s hardware consists of three subsystems: a rotation subsystem, an auto-adjustment subsystem, a skin-stretching subsystem, and a system board. The device can be secured to the forearm with a rubber band. Depending on the range of arm deformation in VR, the hardware stretches the skin to varying extents, simulating the effect of different degrees of arm deformation. Our hardware frames are 3D-printed, as depicted in Fig.  8, using black PLA and orange PETG. The device’s overall weight is 834 grams.

Figure 8:

View Figure

6.1.4 System Board Design.

To enable the operation of four actuators operating at 6V. Please refer to Appendix  B.

Figure 9:

View Figure

6.2 Device Software

The device code runs on Arduino Mega, and our VR applications are developed using the Unity Game engine (2021.3.26f1c1) on an ASUS Vivobook Pro 16X OLED (K6604) PC. The users put on the ArmDeformation along their forearm axis, where the Rings are orthogonal to the forearm axis and define the XY axis. The y-axis is set to point straight up from the forearm. When the device is turned on, it will wait for a command from Unity. By default, the device’s auto-adjusting actuator pulls the racks away from the user’s forearm, keeping the silicone pads distant from the skin. At the same time, the skin-stretching actuators place the silicone pads in the middle of their range of motion, allowing the pads to move in both directions upon command.

Whenever an arm deformation is needed, the Unity application sends the specific type of deformation required to the device. For forearm bending, it sends bending direction and angle; for elongation or shortening, it sends the proportion of length to be deformed. After the device receives the command, if it is a forearm bending, it will use the XY motor to rotate the rings to the set position along its track. Then, in all arm deformation conditions, the auto-adjusting actuators push the rack nearer to the forearm skin, stopping when the microswitch is triggered. After the auto-adjusting actuators have stopped working, the skin-stretching actuators perform operations depending on the type of arm deformation. The skin-stretching actuators move the silicone pad toward the forearm bending direction for the forearm bending. The actuator on the same side of the bending direction moves the silicone pad inward, and the actuator on the opposing side moves the silicone pad outward. The movement range of skin-stretching actuators depends on the forearm bending intensity: a larger bending angle results in more extensive pad displacements. For the forearm elongating and shortening, the skin-stretching actuators drive the silicone pad in sync with the avatar’s forearm-length changing. Once the virtual forearm completes its deformation, the auto-adjustment actuators retract the rack for the next skin-stretching.

6.3 Technical Evaluation

We conducted a technical evaluation of our ArmDeformation prototype to learn about its capabilities and limitations. We were particularly interested in answering four technical questions: (1) What is the normal pressure produced by the auto-adjustment actuator? (2) What is the relationship between the distance the silicone pad moves and the force produced by the skin-stretching actuator? (3) What is the latency of ArmDeformation, from device off to an actual skin-stretching? (4) What is the typical noise produced by Deformation? This section attempts to answer these questions and better inform the system’s design.

6.3.1 Pressure and Force for Skin-Stretching Feedback.

Figure 10:

View Figure

To evaluate the pressure on the skin and force generated by skin-stretching feedback, we built a special experiment setup comprised of a simplified ArmDeformation prototype and an artificial silicon skin placed on a base positioned beneath it, as shown in Fig.  10 (a). The simplified ArmDeformation prototype consists of an auto-adjustment actuator, which generates pressure, and a skin-stretching actuator, which creates a force for skin-stretching feedback, both used the same actuator (MG996R) as the actual prototype. First, we find the relationship between the torque (T) and current (I) of the actuator we used. The torque-current relation of the actuator was determined by having it hold a weight (50g to 550g) on the end of a gear, the distance between the weight and actuator being 250mm, and recording the current draw of the actuator. After conducting tests on 55 different data points, torques ranged from 122.5N · mm to 1347.5N · mm. We fit a linear function to these data points using the least squares method and found that the actuator had a linear relation of T = 1.83I, where T is in N · mm, and I is in mA, as shown in Fig.  10 (b). The line of best fit had an R² value of 0.942. Second, to measure the pressure, We conducted 30 measurements of the current of the auto-adjustment actuator. The current was 203.33mA (SD: 20.90), and the calculated torque was 372.10N · mm (SD: 38.24). In the end, we divided the torque by the gear’s radius (29.5mm) to determine the pressure we applied on the skin. Based on this calculation, the pressure of skin-stretching feedback in our device is 12.61N (SD: 1.30). Third, we measured the current of the skin-stretching actuator to find the force for skin-stretching feedback. Considering the isotropic elastic behavior of the skin  [71], we controlled the actuator rotated to move the silicone pad between 0 to 10mm with 1mm increments and repeated three times. Paired displacement and current values readings were taken in 1mm increments, resulting in 60 data points. We used the least squares method to fit a power function to these data points. Based on this, we can figure out force 0.45N (SD: 0.24) in 1mm to 20.05N (SD: 5.48) in 10mm. Also, based on the result, we calculated the power function between distance traveled (D) and force (F), which is F = 0.02D^2.91, as shown in Fig.  10 (c).

7.1 Procedure

Figure 11:

View Figure

In this study, we implemented a 2AFC (two-alternative forced-choice) test, where participants were tasked with identifying the direction of the test direction—counterclockwise or clockwise—concerning the reference direction within a sequence of two different forearm bending directions [9, 47], which can estimate PSE and JND values from various directions.

The reference forearm bending was set in eight directions at 45° intervals. The test bending direction was positioned ± 20° , ± 45° , and ± 90° relative to these references [9, 47]. Each reference and test bending direction combination was repeated five times, resulting in 240 trials per participant (calculated as 8 reference bending directions × 6 test directions × 5 repetitions).

The order of the reference and test bending directions was randomized to avoid any potential pattern recognition or predictability. Both the reference and test bending directions were displayed for a 5-second duration. Participants were able to answer by selecting either the counterclockwise or clockwise option using the left or right side of the VIVE controller touchpad, respectively, as shown in Fig.  11.

7.2 Participants

We recruited 12 participants (six self-identified females and six self-identified males) aged between 22 and 27 years (M = 24.2, SD = 1.3), all right-handed. Their forearm length (including the hand) ranged from 410 to 450 mm (M = 425, SD = 14.5). Of these participants, 13 reported using VR technology, and six were aware of visuo-haptic illusions. All participants were reimbursed with an amount equivalent to 15 USD in their local currency for their participation. Our Institutional Review Board approved the study.

7.3 Results and Findings

Figure 12:

View Figure

Fig.  12 shows the results from our user evaluations. With angle deviations of ± 20°, ±45°, ±90° , the mean accuracy rates registered at \(88.33\%\) (SD: 0.086), \(92.81\%\) (SD: 0.051), and \(94.58\%\) (SD: 0.055) respectively. This indicates a correlation where a decrease in angle difference leads to a reduction in the rate of correct identification.

The point of subjective equality (PSE) shows the difference between veridical and perceived directions, and JND shows the precision of direction judgments. In determining the PSEs and JNDs across the eight directions, we adopted a methodology consistent with prior research  [9, 47], leveraging a sigmoid equation (f(x) = 1/(1 + e^{− ((x − α)/β})). This equation was fitted to the participants’ clockwise response ratio, with α and β as constants determined through a least squares approach. We used the ratio of the clockwise answer in the ± 20°, ±45°, ±90° test directions to fit the formula to obtain the user’s PSEs and JNDs for the eight reference directions. In Fig.  12 (d), the plotted dots represent the proportions of clockwise responses, while the sigmoid equation delineates the fitted psychometric function. The PSE values are given by a proportion of 50 percent for the psychometric function  [9]. The specific PSE values for the eight directions are illustrated in Fig.  12 (e). The difference between the two angles, where the curves correspond to values of 0.84 and 0.5, determines each JND value. Based on the JND for each participant, we conducted a Friedman Test post-Wilcoxon signed rank test. The result shows significant differences between 0° and 135° (p = 0.003), 0° and 180° (p = 0.007), 0° and 225° (p = 0.005), 0° and 270° (p = 0.036), 0° and 315° (p = 0.016), 45° and 135° (p = 0.004), 45° and 180° (p = 0.014), 45° and 225° (p = 0.005), and 90° and 225° (p = 0.033). The specific JND values for the eight directions are illustrated in Fig.  12 (f).

Our PSE values range from -8.4° to 9.0° and JND values span from 5.64° to 27.97°. Using this data, Immersion can be enhanced by adjusting the visual feedback in VR and controlling the ArmDeformation to match the user’s expectations and perceptions. The PSEs show how users tend to perceive bending directions, so visual cues in VR applications can be fine-tuned to make them a more natural and immersive experience. We also can use the value of JND to overcome the latency limitations of ArmDeformation, e.g., ArmDeformation doesn’t need to rotate when the difference in the angle of the direction in which the user’s forearms are bending is less than the corresponding JND value.

8.1 Procedure

Figure 13:

View Figure

The first application we explored was a virtual boxing game. Participants can control their forearm elongating by pressing the trigger button on the VIVE controller, allowing them to interact with the sandbags. When the hand hits the sandbag or reaches its maximum distance, there is a rebound recovery process for the forearm. In this application, since we don’t need to change the position of the skin-stretching subsystem, the skin-stretching feedback will not have a delay.

The next application provides users with an experience that controls the movement of a tentacle. The skin-stretching feature of ArmDeformation allows users to feel the various bending of the tentacle, which they can control by wearing a VIVE tracker on the back of their hands and forearms next to the elbow and bending it at any angle. Through a first-person perspective, users can fully immerse themselves in the experience of controlling the tentacle’s movements in both horizontal and vertical directions. They can easily switch between these directions by pressing the trigger button on the VIVE controller, making the experience even more engaging. In this application, if we do not need to change the bending direction, the skin-stretching feedback will not have a delay.

In each application, the user wore the ArmDeformation in their dominant hand while wearing the VIVE HMD. Participants were asked to experience the two applications in two conditions (visual-only, visual + haptic) for at least 3 minutes. Following each condition, the participants evaluated the level of realism and enjoyment of their experience, as has been done in previous research [40, 55, 95], by filling out a paper questionnaire with 7-point scale Likert questions. The experiment ended with a post-hoc interview. The experiment took about 30 minutes to complete.

Our study utilized a within-subject design with a single modality factor: haptic (our device) versus no-haptic experience (baseline condition). During the haptic condition, participants were exposed to VR applications with skin-stretching feedback generated through the ArmDeformation. Conversely, during the no-haptic condition, participants experienced VR applications without haptic feedback, relying solely on visual feedback.

8.2 Participants

We recruited 16 participants (six self-identified females and ten self-identified males) aged between 21 and 27 years (M = 24.4, SD = 1.9), all right-handed. Their forearm length (including the hand) ranged from 400 to 490 mm (M = 433.7, SD = 27.2). Of these participants, 14 reported being familiar with VR technology, and ten were aware of visuo-haptic illusions. All participants were reimbursed with an amount equivalent to 10 USD in their local currency for their participation. Our Institutional Review Board approved the study.

8.3 Results and Findings

Figure 14:

View Figure

Fig.  14 provides the subjective evaluations of realism and enjoyment across conditions. We subjected the results to the Wilcoxon signed-rank test. The evaluations highlighted that, when compared to the visual-only condition, the visual with haptic condition was significantly better in terms of realism for both boxing (Z = −3.559, p = 0.0003) and tentacles scenarios (Z = −3.035, p = 0.002). The same superiority in the visual with haptic condition was evident for enjoyment, with significant differences observed for boxing (Z = −3.211, p = 0.001) and tentacles scenarios (Z = −3.559, p = 0.001). These results suggest that our system can significantly improve the realism and enjoyment of VR experiences.

8.4 Qualitative Findings

Qualitative results from interviews with our participants further confirmed our previous findings. All participants agreed that using ArmDeformation felt more realistic and enjoyable. As a result, it was also more immersive. For example, P8 said, "It was clear to feel the difference between visual-only and visual with haptic conditions, and it felt as if you were reaching a sandbag yourself with the haptic feedback, rather than moving an abstract thing inside the computer.", and others described it as "novel" (P4) and "interesting" (P12). When asked about their favorite part of the experiment, for the tentacle scenario, P5 said, "In the tentacle scenario, I would feel that the angle of the bending of the tentacle could be matched with the movement of the hand and feel that my hand was the tentacle. It felt the movements were pretty synchronized and realistic." Participants also commented on the tentacle bending solution and were satisfied with it, with P11 saying, "It’s hard to feel that the tentacle is my hand, but when the tentacle is moving, it still feels quite real, it can give me a feeling that the tentacle is soft, not a rigid body like the forearm."; and for the application scenario of boxing, P12 said, "When the device works, it will feel like my forearm is elongating, especially when the forearm hits far away, the device will give more strength, and it will feel like much realism.", P9 said, "The added haptic feedback would feel like the forearm is exerting force." However, some participants suggested that the forearm elongating length is a bit unrealistic; P8 said, "There will be a certain gap between the forearm elongating to a certain length and the reality." P11 said, "The forearm elongating is too long. It feels a bit unrealistic and like an animation." But they still felt more realistic than the visual condition. At the same time, some participants also suggested application scenarios, "I can feel the difference between haptic and no haptic. In more reality-oriented scenarios, I prefer no haptic; in scenarios that require fine manipulation and the forearm or tentacle in the VR will move along with the movement in reality, I prefer haptic, such as the movement of the tentacles."

When asked which application scenario participants preferred, most participants (P1, P3, P4, P5, P6, P10, P11, P12, P14, P16) found the boxing application scenario to be more immersive than the tentacle application scenario; this is not surprising as many people are not used to non-human avatars. P6 and P16 noted that it was more realistic and enjoyable because "there were things in the scenario that were able to be interacted "; P4 and P12 mainly appreciated the feeling of the forearm elongating, "It felt more realistic, it felt like the forearm elongating that long.". As for the scenarios where tentacles are used, P7 and P13 state that "the tentacle ideas are better, and the feeling of the tentacle movement is exciting."

Finally, when asked what needs to be further improved, all participants hit on the fact that they wish the device could just be a little bit lighter, even though the current experience is satisfactory. They commented that if the device was more lightweight, it could be applied to more parts of the body, e.g., P1 said, "If the hardware could be lighter, it could be placed in a VR gym application scenario, where the device would be placed on different parts of the body such as the forearms, palms, thighs, etc., and it could be used to stimulate people through skin-stretching to feel the muscle in the feeling of movement." P6 said, "I would be more willing to use the device if it was lighter, and it would be better if it could be made into a glove or something like that, where you can feel the force feedback on the back of your hand and the palm of your hand.". At the same time, some participants pointed out that it would be nice if the device could be immobilized, P7 "Preferably in scenarios where the hardware can be immobilized, such as when using an exoskeleton mech." Some participants (P1, P13) also noted that controlling forearm elongating in boxing scenarios would be nice through forearm movements, "It would be more natural if body movements triggered it." Finally, many participants (P5, P6, P7, P13, P16) noted that having more things to interact with would be nice, "It feels like more things could be added to the scene that can be interacted with."

9.1 Reflections on Studies

Our research demonstrated the positive impact of using body deformation illusions in VR to enhance user body ownership and immersion. Additionally, we found that haptic feedback, explicitly skin-stretching, can further enhance body ownership. In the length adjustment study, we aimed to determine the minimal number of actuators required to induce forearm elongating and shortening illusions convincingly. Our findings indicate that using only two actuators is sufficient to generate these illusions effectively, and it is significantly better than the visual-only condition. This diverges from existing work, such as Gum-gum shooting  [93], which shows the efficacy of skin-stretching techniques but is limited to forearm elongating illusions and lacks empirical investigation into the optimal number of actuators for inducing such illusions. Additionally, our Perception Study 2 result suggests synchronous actuator movements provide a more convincing illusion than using a single actuator or visual feedback alone.

The results of Study 1 were a surprise. The skin-stretch stimuli represent the elongating or shortening of the forearm’s skin due to the deformation. As such, we expect that two locations along the forearm (4 actuators) may render this virtual scaling stronger than one location on the forearm. The result (2 actuators) enabled us to design a lighter, smaller device that requires less power without sacrificing the immersive experience. These findings provide a solid foundation for future research on arm deformation illusions generated through skin-stretching of the forearm.

These insights were integrated into the motion design of our final wearable device, ArmDeformation, which utilizes two skin-stretching actuators that operate synchronously during arm deformation illusion. Furthermore, while previous research have focused on the thresholds for forearm elongating  [46] or practical techniques  [7], our work expands this by identifying angular thresholds for forearm bending illusion. Specifically, our results show that angles of up to 49.4830° for upward bending and 50.5061° for downward bending do not adversely impact the user’s sense of body ownership. This empirical guideline informed our software development, leading us to set a conservative maximum forearm bending angle of 45° in our applications. Such new flexibility can be used in the future to examine new ways of interactions, where small physical movements of the limb may be mapped to a richer set of virtual limb actions, and using haptic feedback to control them better or sense them when they are not of the visual sight.

Our experimental evaluation confirms that ArmDeformation notably augments user immersion in VR. Unlike prior research  [47, 88] that centered on the PSE and JND for haptic perception, we aimed to identify the visual PSE and JND for forearm bending directions that users could reliably perceive. Our JND results indicate that users can distinguish forearm bending directions between 5.64° and 27.97° , and our PSE result shows the PSE values range from -8.4° to 9.0°. These findings offer valuable design considerations, enabling us to employ consistent forearm bending direction within this perceptual range, optimizing visual effects without compromising user experience. To assess the broader implications of our device, we developed possible applications and conducted an immersive study focusing on enhancing realism and enjoyment. The study results show that ArmDeformation significantly elevates realism and enjoyment of the superpower scenario. As Fig.  14 shows, using ArmDeformation is considerably preferred by users over visual-only conditions. This study emphasizes the effectiveness of skin-stretching methods in producing illusions of arm deformation, whereas Gum-gum shooting  [93] lacks supporting data.

9.2 Skin-stretching Feedback for Arm Deformation Illusion

Humans have a limited range of motion when it comes to rotating their forearms. They can only rotate them within a range of 104.54 ± 11.09 degrees  [70], which limits their ability to express the bending of their forearm. However, our research overcomes this limitation using alternate morphologies in body transformations  [6] in VR, which is forearm deformation. Our study results show that this approach is consistent with earlier research studies  [10, 75, 87], which aimed to enhance users’ joy in VR through different body deformations.

In previous studies  [45, 60, 67], it has been introduced that when sensory information is integrated, users tend to have a stronger sense of embodiment with their avatars, aligning with our findings. This integration, particularly the addition of haptic feedback, is known to lessen the cognitive load on users  [37]. Our research aligns with this by employing a dot-shaped silicone pad for skin-stretching feedback, enhancing the user’s sense of body ownership. This approach differs from Gum-gum shooting’s  [93] method of using bands for skin-stretching to simulate elongation. Instead, we use a dot-shaped pad that stretches the skin forward and backward, creating an immersive illusion of forearm bending. This method was also chosen over other techniques like EMS  [54, 55] or vibration  [33, 57] to deliver a strong and bidirectional force feedback at a single skin point. Our study’s results indicate that our skin-stretching design effectively simulates the sensation of forearm bending.

Users often struggle with full immersion in VR when embodying avatars with fantastical features, like the superhuman abilities of Rubberman  [15]. To overcome this, our research introduced an immersive arm deformation illusion by focusing on creating, differing from, and aligning with previous research that adjusts body parts’ length  [46, 50, 66, 93] or overall body size  [5, 10, 64, 75, 87]. Our initial studies, Studies 1 and 2, have proven that combining visual and haptic feedback enhances the sense of body ownership during these arm deformation experiences. Following the insights from Study 3, we can more convincingly generate the sensation of forearm deformation in VR, ensuring that users perceive the virtual forearm as their own. This approach allows us to present the illusion of forearm deformation without compromising the user’s sense of body ownership, such as deepening immersion into superhuman capabilities in VR. According to Study 5’s findings, users can have a realistic and enjoyable experience in VR, using ArmDeformation to elongate, shorten, or bend their forearms, thereby achieving a convincing and immersive arm deformation illusion.

9.3 Possible Applications

Figure 15:

View Figure

Beyond the earlier-mentioned possible applications, we’ve thought of two more, as Fig.  15 shows. The first is a baseball game where our device makes players superhumans, and the haptic feedback enables them to control their rubber-like forearms. As catchers, players can sway their arms to catch balls they could not have done before, and as pitchers, they can control this bending using their wrists and use the new virtual elasticity of their limbs to generate super-powerful swings. This match-up is an example of adding new physical abilities that were impossible in real life and can generate new possibilities and interest in existing games. Our second idea is a "remote click" application for intelligent homes (see Fig.  15 right). Here, users control home devices like lights and TVs by elongating their forearms in VR and even bending them across a corner.

Besides entertainment, the ability to map users’ small motions to large and even physically impossible motions in VR is an essential capability for applications such as productivity and accessibility  [6, 53, 86]. It enables users to do large actions in the visual space without getting exhausted in the physical world or to level the virtual plane field for users limited in the extent and range of their physical motions in the real world. Haptic feedback that renders the new capabilities to the users is essential to learn new dexterities and better control, as well as feedback to the user when the limbs are not in the focus of the visual field  [3, 82, 90].

9.4 Limitations and Future Works

While we have tried to make our user research as good as possible, it still inevitably has limitations. In Studies 1 and 2, we considered the skin-stretching design after determining the number of actuators, but the skin-stretching design may also have different impacts on body ownership with different numbers of actuators. In future work, we will recruit more people to conduct more complete experiments that consider different skin-stretching designs with different numbers of actuators. In Study 3, we only considered thresholds for forearm bending upward and downward and did not consider thresholds for forearm bending in other directions. In future work, we will conduct more detailed experiments to establish complete thresholds for other bending directions. In Study 4, we use the same approach as in the previous research  [9, 47], using an estimation method to assess the JND and PSE values, which may have discrepancies between the actual values. In future research, we will recruit more participants for the JND and PSE study and use more precise measurement methods. In Study 5, we only consider realism and enjoyment. In future work, we can consider more dimensions, such as long-term usability and cognitive load.

ArmDeformation can enhance the realism and enjoyment of VR experiences, as demonstrated in User Study 5. However, it still has some limitations due to its prototype nature. Its weight can be reduced by using lighter materials like carbon fiber or LW-PLA, and its design can be more ergonomic and comfortable. The motors used can also be more lightweight, such as Dynamixel XL-320, and DF9GMS. This could also help reduce the size of the overall wearable.

Rotating the actuators in a new direction may delay the identification of a wish to bend the forearm and the actual skin-stretching rendering. This latency may be reduced by rotating the actuator prehand based on anticipating possible rotation directions but at the cost of increasing power consumption. The usage of predictive algorithms can also reduce average delay. Another limitation of the current implementation is using a fixed bending direction once the bending has begun. While the dominance of the visual sense can be used to show some rotation of the bending direction, enabling this capability may become a subject of future research.

We found that using natural anchor points of the skin (e.g., the elbow) might be enough to generate a skin-stretching effect that fits the arm deformation illusion. Yet, it ignores other natural anchor points on the other end of the arm, which might have generated an opposing stretch feeling. In future work, we will explore additional natural anchor points for arm deformation illusion creation. The dynamic range of skin-stretching rendering is limited. We intend to explore further ways to better use the stimuli, from understanding the stretch JND to nonlinear mapping from bending angles to haptic-rendered stimuli.

In future work, we will explore the advantages of integrating different haptic feedback methods, like EMS  [54, 55] and vibration  [33, 57], with the skin-stretching technique to enhance body ownership. Moreover, we plan to investigate more haptic pattern designs that can induce various arm deformation illusions. Unlike the Gum-gum shooting  [93], ArmDeformation lacks the function of changing the center of gravity. Therefore, future work will also explore the possibility of incorporating changes to the center of gravity along with our skin-stretching technique. Additionally, we will use faster servo motors to optimize mapping speed between skin-stretching and visual feedback. Other potential applications of arm deformation will be explored, such as retargeting and health.

Finally, the focus of this work was to examine the haptic rendering of forearm bending, and we used a simple interaction scheme, using a VIVE Controller to let the user control the bending. We intend to develop a more natural way of controlling arm deformation in VR, for example, using the wrist  [23] and palm gestures  [24, 26]. We consider a technique that is similar to FingerMapper  [86]. FingerMapper mapped small-scale finger movements to the virtual arm and hand. Referring to this, we explore mapping the movement of the user’s wrist to the degree of bending in the forearm.

A APPENDIX: GEOMETRIC BENDING OF THE HAND MODEL

In VR, we utilized a specific model to simulate forearm bending. Suppose the forearm’s length is L_arm. The forearm gradually assumes a semi-circular shape as the bending angle α increases. In Unity, we calculate the coordinates of the vertices of the bent forearm. The first step is to determine the location of the calculation point projection on the forearm centerline. Subsequently, by introducing suitable variables, we can accurately compute the coordinates of the vertices on any side of the centerline.

Relying on the forearm bending model previously mentioned, we can calculate the coordinates of the changed forearm centerline. In this context, R is the semicircle’s radius, d_arm represents the ratio of a projection point on the centerline’s distance from the base to the length of the forearm in its initial state, β denotes the point’s bending angle in the model, and γ indicates the direction of the forearm’s bend (the plane is perpendicular to the forearm, with 0° directly above it, the y-axis is aligned with the forearm’s extension, while the x-axis aligns with 0°): (2) \(\begin{align} R&=\frac{L_{\text{arm}}}{\alpha } \end{align}\) (3) \(\begin{align} \beta &=\alpha * d_{\text{arm}} \end{align}\) (4) \(\begin{align} x&=R *(1-\cos \beta) * \sin (\gamma) \end{align}\) (5) \(\begin{align} y&=R * \sin \beta \end{align}\) (6) \(\begin{align} z&=R *(1-\cos \beta) * \cos (\gamma) \end{align}\) After calculating the forearm centerline coordinates, it becomes necessary to compute the offset of the vertex in the XZ plane. In this equation, x_o, z_o represent the vertex’s initial coordinates, while L_offset denotes the distance of the vertex from the centerline in its initial state, also taking into account the vertex’s direction γ_this relative to the x-axis: (7) \(\begin{align} L_{\text{offset }}&=\sqrt [2]{x_o^2+z_o^2 } \end{align}\) (8) \(\begin{align} \gamma _{\text{relative }}&=\arctan \frac{x_o}{z_o}-\gamma \end{align}\) (9) \(\begin{align} x_{\text{relative }}&=L_{\text{offset }} * \sin \left(\gamma _{\text{relative }}\right) \end{align}\) (10) \(\begin{align} z_{\text{relative }}&=L_{\text{offset }} * \cos \left(\gamma _{\text{relative }}\right) * \cos (\beta) \end{align}\) (11) \(\begin{align} x_a&=x_{\text{relative }} * \cos (\gamma)+z_{\text{relative }} * \sin (\gamma) \end{align}\) (12) \(\begin{align} z_a&=-x_{\text{relative }} * \sin (\gamma)+z_{\text{relative }} * \cos (\gamma) \end{align}\) Calculates the offset of the vertex in the y-direction of the point: (13) \(\begin{align} y_b=-L_{\text{offset }} * \sin (\beta) * \cos \left(\gamma _{\text{relative}}\right) \end{align}\) Finally, we can add up the individual quantities to get the coordinates of each vertex(x + x_a, y + y_b, z + z_a) of the forearm.

To bend the forearm model while keeping the hand model tangent to it, we calculate the position of each vertex of the hand using the following formula. Suppose the hand’s length is L_hand, and d_hand represents the ratio of a point on the centerline’s distance from the base to the length of the hand in its initial state. (14) \(\begin{align} x_{\text{hand}} &= R * (1 - \cos \alpha) * \sin \gamma + L_{hand} * d_{hand} * \cos {(\cfrac{\pi }{2} - \alpha)} * \sin \gamma \end{align}\) (15) \(\begin{align} y_{\text{hand}} &= R * \sin \alpha + L_{hand} * d_{hand} * \sin {(\cfrac{\pi }{2} - \alpha)} \end{align}\) (16) \(\begin{align} z_{\text{hand}} &= R * (1 - \cos \alpha) * \cos \gamma + L_{hand} * d_{hand} * \cos {(\cfrac{\pi }{2} - \alpha)} * \cos \gamma \end{align}\) The offsets of the hand’s vertices in the x, y, and z directions are the x_a, y_b, and z_a found when d_arm = 1. The hand’s computed vertex position is therefore (x_hand + x_a, y_hand + y_b, z_hand + z_a)

B APPENDIX: SYSTEM BOARD DESIGN

We opted for the LM7806 voltage regulator to set up a simple circuit that converts 12V to 6V. According to the MG996R datasheet, under a 6V working condition, their current consumption ranges from 500mA to 900mA. For four such actuators, the cumulative current requirement is between 2A and 3.6A. However, the maximum output current of the LM7806 is just 1A, which does not meet the requirement. Additionally, the LM7806 generates significant heat during operation, so we must consider heat dissipation and cost. Therefore, we designed an external circuit for the LM7806. We employed a PNP transistor (TIP2955) to amplify the current. According to the TIP2955 datasheet, when the voltage between the collector and emitter is 4V, the minimum amplification factor is 20, and the collector can output a current of 4A. Given that the emitter of the transistor is connected to the 12V power source and the collector is connected to the 6V output of the LM7806, there’s a voltage difference of 6V. Our calculations determined that a 0.5-ohm resistor connected to the emitter would suffice. However, to be cautious, we opted for a 1-ohm resistor. Even though the amplification factor didn’t reach 20, our tests confirmed that the circuit can still drive all four servos synchronously.

For the DC motor, we selected the motor drive chip (L293D) to drive it. Moreover, the enable pin of the L293D is connected to the PWM pin 3 of the Arduino Mega, allowing us to control the motor’s rotation speed. The DC motor operates at 12V, so the VS pin of the L293D is directly connected to the 12V power source. Additionally, our PCB has two connections to microswitches (1A 125V AC). When the microswitch is not pressed, the Mega reads a high voltage level; when pressed, it reads a low level. These microswitches detect if the skin-stretching subsystem is touching the user’s skin.