SplitBody: Reducing Mental Workload while Multitasking via Muscle Stimulation

Techniques like electrical muscle stimulation (EMS) offer promise in assisting physical tasks by automating movements, e.g., shaking a spray-can or tapping a button. However, existing actuation systems improve the performance of a task that users are already focusing on (e.g., users are already focused on using the spray-can). Instead, we investigate whether these interactive-actuation systems (e.g., EMS) offer any benefits if they automate a task that happens in the background of the user's focus. Thus, we explored whether automating a repetitive movement via EMS would reduce mental workload while users perform parallel tasks (e.g., focusing on writing an essay while EMS stirs a pot of soup). In our study, participants performed a cognitively-demanding multitask aided by EMS (SplitBody condition) or performed by themselves (baseline). We found that with SplitBody performance increased (35% on both tasks, 18% on the non-EMS-automated task), physical-demand decreased (31%), and mental-workload decreased (26%).

Figure 1: (a) This user finds themselves having to split their attention (dashed green arrow depicts their split attention) between two sub-tasks, continuously stirring the pot to make caramel (a repetitive muscle movement) and writing an essay (a cognitively-demanding task)-multitasking is hard and even a simple repetitive muscle task detracts an untrained user from devoting more cognitive resources to the competing task.To explore this space, we (b) propose using electrical muscle stimulation (EMS) to "split" the user's body and allow the EMS (lightning icon depicts the electrical stimulation) to automate the simple & repetitive muscle movements while focusing on writing (depicted by the solid green arrow).

ABSTRACT
Techniques like electrical muscle stimulation (EMS) offer promise in assisting physical tasks by automating movements, e.g., shaking a spray-can or tapping a button.However, existing actuation systems improve the performance of a task that users are already focusing on (e.g., users are already focused on using the spraycan).Instead, we investigate whether these interactive-actuation systems (e.g., EMS) offer any benefits if they automate a task that happens in the background of the user's focus.Thus, we explored whether automating a repetitive movement via EMS would reduce mental workload while users perform parallel tasks (e.g., focusing on writing an essay while EMS stirs a pot of soup).In our study, participants performed a cognitively-demanding multitask aided by EMS (SplitBody condition) or performed by themselves (baseline).We found that with SplitBody performance increased (35% on both tasks, 18% on the non-EMS-automated task), physical-demand decreased (31%), and mental-workload decreased (26%).

INTRODUCTION
Techniques such as electrical muscle stimulation (EMS) offer promise for assisting users with physical tasks.These interactive systems do this by automating entire movements, e.g., shaking a spray-can [47], tapping a button on a touchscreen [29] or playing a musical instrument [87].However, it is key to note that the majority of these actuation systems improve the performance of a task that users are already focusing on (e.g., users are already focused on using the spray-can in [47] or already attempting to press the button in [29]).In other words, body-actuation driven by interactive systems (e.g., EMS) happens in the foreground of the user's main focus of attention, in which the interactive system assists the user in completing the same task the user is also attending to.
Instead, we investigate whether these interactive actuation systems (e.g., EMS) offer any interactive benefits if they automate a task that the user is not attending to.In other words, we explore a novel space in which the body actuation happens in the background, out of the user's main focus of attention, enabling users to potentially attend to another task in parallel-this would enable new forms of physical multitasking.
However, much is unknown about body-actuating systems.When it comes to EMS, only recently have researchers found that actuating the user's muscles can also decrease the user's sense of agency [29,30,85] or even distract users (e.g., EMS causes a tingling that can distract [39,44,65,74,89]).Thus, while actuation systems can automate simple & repetitive gestures, it has not been studied whether their limitations (e.g., tingling or loss of agency) might prove detrimental to task performance by distracting users.
To shed light on this, we conducted a study where participants performed a cognitively-demanding multitask, in which both their hands performed parallel tasks: a repetitive movement task and a cognitive task.They performed these tasks twice, once with the movement task aided by EMS on one hand (a condition we call SplitBody) and another entirely by themselves (baseline).
We found that with SplitBody, participants reported less physicaldemand (decrease of 31%) and less mental-demand (decrease of 26%) than when performing the task by themselves.Moreover, the performance increased by 35% (averaged over both tasks), including the task that was not automated by EMS, which increased by 18%.This suggested that, with SplitBody, participants were able to free-up cognitive resources that they then allocated to this task.Moreover, accounts of their experience suggested they felt less overwhelmed since they could just focus on one task-the non-automated cognitive task with SplitBody while not being distracted by the EMS moving their arm involuntarily.

RELATED WORK
Our work builds on interactive systems based on EMS that assist users by electrically actuating their limbs.While EMS is not the only class of haptic actuators capable of displacing limbs involuntarily, we focus on it due to its inherent wearability when compared to mechanical actuators [8-10, 18, 24, 57, 66]-EMS does not require heavy & cumbersome power supplies (e.g., batteries or air compressors).

EMS is always in the foreground
EMS systems are typically used to aid users in tasks they are already focused on.For instance, in the seminal PossessedHand [87], EMS plays the next musical note for a user who is already seated & playing this musical instrument.Similarly, in Affordance++ [47], EMS shakes a spray-can that the user is already focused on using.As such, designers & engineers tend to position EMS in the foreground of the user's attention by using the stimulation to tackle primary tasks.This approach proved successful in that it sparked new usages of EMS for interactive contexts.Thus, it inspired us to propose a new design position for EMS in the background, i.e., assisting the user by automating repetitive tasks, and leaving the user's attention to focus more on more challenging concurrent tasks.
To the best of our knowledge, this has not been systematically studied.The idea that comes closest and from which we draw inspiration is Pedestrian Cruise Control [73], an EMS system that turns the user's legs.While Pfeiffer et al. envisioned their system going as far as to help users by steering them automatically, they only tested it in a study where participants were not engaged in a multitasking scenario; in fact, participants who were being guided by the system to walk in a park were asked "to pay attention to any obstacles, (. ..) and to stop or circumvent these as necessary"; thus, the EMS of [73], much like prior work, was acting in the foreground of the user's attention.

Mixed agency while interacting with "integrated" devices
There is an emerging body of literature exploring the concept of mixed agency [55] between a user and their device [54,55].While EMS provides one of the most extreme case-studies for this concept [11,15,29,30,33,46,51,55,81,82,85], others have also started to discuss this for the case of force-feedback devices [12].Central to the frameworks of thought in this area (e.g., humancomputer integration [55] or [12]) are two dimensions: sense of body-ownership ("this is my body") and sense of agency ("this is my action") [12,54].In Mueller et al.'s taxonomy [54], most EMS interfaces score low on agency since users do not initiate the action that the EMS carried out.Conversely, EMS interfaces score high in ownership, since it is the user's body that carries out the actions [54].The authors also emphasize that if the EMS system does not control the entire body, some agency remains [54]; this is also the case in SplitBody, as our users were always in control of other upper joints of the actuated extremity (e.g., they controlled their shoulder while their hand was EMS-automated).Taken together, this allows us to denote SplitBody as a type of EMS interface with high body-ownership but low agency.
Finally, it has been argued that these mixed agency devices introduce new design spaces that foster playful experiences [46,54,71], creativity [11,13,94], and even new forms of productivity [69,83], to which we believe SplitBody contributes to with its multitasking perspective (i.e., design spaces with concurrent physical tasks).

Evidence of lower performance in physical multitasking
It is well-known that multitasking comes at a performance cost compared to only focusing on a single task.Neuroscientists and psychologists have correlated this cost to the limited cognitive resources available [4,43] and the operations the brain undertakes when presented with a task: for each task, the brain has to ingest the information, process it, make a decision, and respond with movements.Therefore, when presented with concurrent tasks, the brain uses different strategies [27,70], such as processing one task before moving to the next one (causing a bottleneck) [6] or reducing the capacity to process tasks in parallel [59,90]-both strategies increase completion time [53,76] and lower performance [67].To improve physical multitasking, researchers have looked at different interfaces, such as adding haptics.

Exploring haptics for mental workload reduction
Adding haptics allows the conveying of additional information to the user [1,3,26,72].Researchers have found that haptic cues can improve the performance of single tasks [1,58].More recent studies have also shown that similar results can be seen while multitasking, lowering mental workload.For example, Zhou et al. [93] showed that surgeons using a haptic simulator performed better against a no-haptics simulator while concurrently answering arithmetic problems.Leung et al. [40] also found that adding haptic-feedback to a touchscreen improved the response time when undergoing a concurrent auditory-task, but no performance improvement was found.Moreover, Haghighi et al. [20] investigated the ability to recognize haptic cues with vibration while performing a cognitive task (1-back) and found that some parameters improved response time, but none improved performance.These haptic systems provide only haptic cues (e.g., vibrations or resistance) but do not move the body.Thus, users still need to focus on executing the movements required to perform all concurrent tasks.

USER STUDY: DOES EMS REDUCE MENTAL WORKLOAD DURING PHYSICAL MULTITASKING?
The goal of our study was to evaluate whether the use of an interactive-actuation system (in this case, EMS) would reduce mental workload while users perform parallel tasks.To this end, we designed a multitasking study in which participants were asked to perform a multitask: aided by EMS (our SplitBody condition) or performed by themselves (baseline).Our study was approved by our ethics review board (IRB21-1158).

Hypotheses
Our main hypothesis (H1) was that the SplitBody's ability to automate one of the tasks would result in a decreased mental workload, when compared to the baseline; as such, we utilized the NASA-TLX questionnaire [21].Our secondary hypothesis (H2) was that this reduced mental workload (i.e., if H1 was true), we would observe an increased task performance; as such, we measured task accuracy and response time.A corollary of the previous hypothesis was that both performances of each sub-task should increase, i.e., (H2.1) increased performance of the task automated by EMS (movement task); and (H2.2) increased performance of the voluntary task (cognitive task).

Study procedure
Study design.Our multitask was based on two standardized designs: (1) a cognitive task-the n-back task [32], a popular cognitive load test [23,25,28,68]; and (2) a movement task-a repetitive sequence of movement often used to analyze cognitive load [41,42,64]; except participants were requested to perform these two tasks simultaneously-resulting in a challenging multitask, similar to the one depicted in our Figure 1.
Participants.We recruited 12 participants (six female and six male; average age=27.4years old; SD=8.0).No participant reported any motor impairment.Participants were compensated with 20 USD for their time.
Conditions.Participants performed the multitask twice, once per condition (condition order counterbalanced across participants), as depicted in Figure 2.During baseline, participants performed the multitask by themselves.During SplitBody, participants performed the cognitive task, while EMS performed the movement task.
Cognitive task (dominant-hand).The objective of our cognitive task was to keep observing a sequence of letters on a screen, shown one at a time, and respond if the current letter was previously shown-n-back task [28].We utilized an N=2, i.e., participants indicated if the current letter was shown two letters ago (2-back).If it was a 2-back, participants were asked to press the left arrow.Conversely, if it was not, press the right arrow.If the participants failed to press a key or if two keys were pressed, their response would be considered an error.A total of 32 letters were presented for 500 ms each at 2500 ms intervals.Two predefined sequences of letters were generated from the following eight visually distinct letters: B, F, K, H, M, Q, R, X where each letter appeared exactly four times in both sequences.Of the 30 responses (first two stimuli do not have n-backs), 10 were n-backs, and 20 were not.Furthermore, the difficulties of both sequences were equalized by featuring three types of n-backs at equal numbers (here illustrated with A, B, C for explanation purposes): "A, B, A" was shown four times, "A, B, A, B" was shown twice and "A, B, A, C, A" shown once per sequence.
Movement task (non-dominant hand).The goal during the movement task was to perform a sequence of hand-gestures for as long as the trial lasted, as depicted in Figure 2 (c): up, down, left, right, down, up, right, and left at a constant tempo of 50 BPM.In the baseline condition, an audible metronome was heard (no metronome was used in the SplitBody condition, as EMS already provides a temporal cue).A complete sequence was considered valid if the participant's hand performed all eight gestures in the correct order (i.e., each gesture was performed before the end of the current beat, and two gestures were not performed within the same beat).
Combined task design.Performing these two tasks simultaneously is challenging.Thus, we designed the combined task to start incrementally, i.e., participants first started the movement task (two complete rounds of the sequence), and only then did the cognitive task start.This was beneficial, especially for the baseline condition, allowing participants to get "a feel" for the movement before adding the cognitive task.
Apparatus.Movements were tracked using a VIVE Tracker 3.0 attached to the hand.An additional RGB camera was used to record trials and transcribe post-interviews.In the SplitBody condition, we utilized a medical-grade muscle stimulator (HASOMED RehaMove3 [75]).The stimulator interfaced with the n-back software, which we implemented in Python via the RehaMove3's library 1 .
EMS parameters & calibration.We attached four pairs of electrodes to participants' muscles at the: palmaris longus for wrist flexion (right gesture), carpi radialis longus for wrist extension (left), biceps (up), and triceps (down).Each participant was calibrated so that the EMS parameters could robustly actuate each gesture.First, we determined the stimulation intensity (i.e., current in mA) for each muscle by starting at 0mA and a pulse-width of 300 `s and increasing by steps of 1 mA until a full & repeatable contraction was observed while also being comfortable to the participant (no pain, cramps, etc.).The pulse frequency was fixed at 100Hz.This process was repeated for all muscles following an anatomical guide.

Trial design & metrics
Warmup.Prior to the tasks, participants were introduced to EMS by having their hands actuated at a constant speed of 30 BPM for three minutes while, simultaneously, the experimenter explained the n-back.After this explanation was completed, participants performed the multitask (once per condition, order counterbalanced). Trial.
Participants were asked to score as high as possible on both tasks.A trial started with a visual countdown and ended when the n-back letters finished.At the end of a trial, participants complete an unweighted NASA-TLX questionnaire.Then, participants were invited to provide feedback on what they just experienced.
Performance metrics.(1) Movement task performance was scored by the number of correctly performed sequences divided by the maximum number possible during a trial, which was ten full sequences.Moreover, we also recorded the response time of each movement according to the expected 50 BPM beats.(2) Cognitive task performance (n-back) was scored by the correct number of answers over the total number of letters (30).Moreover, we also recorded the response time.Finally, the NASA-TLX was scored by averaging equally the six metrics (unweighted TLX)-the higher a TLX score is, the higher the perceived workload was.

Results
Movement task performance.We analyze the movement task performance.As our data did not follow a normal distribution using the Shapiro-Wilk test, we conducted a Mann-Whitney U-test.We found a significant difference (p<0.005) between movement task performance of SplitBody (M=79%, SD=26%) and baseline (M=28%, SD=21%).Results suggest that movement performance was  increased 2.5x with SplitBody, as depicted in Figure 3 (b).Note that this was expected since we calibrated the EMS to be robust and it was automating the task.This result supports our H2.1 (i.e., increased performance of the EMS-automated task).
Cognitive task performance.Most relevant to our H1, we analyze the cognitive task performance.As our data did not follow a normal distribution using the Shapiro-Wilk test, we conducted a Mann-Whitney U-test.We found significant difference (p<0.05) between cognitive task performance of SplitBody (M=78%, SD=19%) and baseline (M=60%, SD=23%).These results suggest that cognitive task performance was increased by 1.3x with SplitBody, as depicted in Figure 3 (c).Average wrong-answers decreased by 1.6x with SplitBody (M=13%, SD=6%) compared to baseline (M=21%, SD=9%).Similarly, average no-answers (failed to press either key as a response, potentially caused by mental overload) decreased by 2x with SplitBody (M=9%, SD=19%) compared to baseline (M=19%, SD=24%)-These results are highly supportive of our H1, this task was performed by participants unassisted (no EMS), suggesting that the gain was due to the SplitBody's assistance of the background task.This also further supports our H2.2 (i.e., increased performance of non-automated).Critically, this increase in performance on the cognitive task shows that despite EMS' shortcomings (e.g., EMS can distract with its tingling sensation [39,44,65,74,89]), it provides a benefit when automating another demanding task.
Overall performance.We analyze the overall multitasking performance, i.e., the average of both tasks.As our data followed a normal distribution using the Shapiro-Wilk test, we conducted a two-tailed paired t-test.We found a significant difference (t(11)=5.6,p<0.0005) between the multitask performance of SplitBody (M=78%, SD=13%) and baseline (M=44%, SD=17%).Results suggest that multitasking performance was almost doubled with the Split-Body, as depicted in Figure 3 (a).
Workload (NASA TLX). Figure 4 depicts our results from the NASA-TLX unweighted questionnaire (higher values indicate higher perceived workload).All comparisons below are Bonferroni corrected (U adjusted = 0.0083).
Response time.Figure 5 depicts participants' response time, which, as we will analyze, we found to be statistically faster with SplitBody than with the baseline condition.

Participants' experiences
Perceived improvement.Ten (out of 12) participants expressed that they perceived performing better with SplitBody (P1, P2, P4, P5, P6, P7, P8, P9, P10, P12).For instance, P12 stated, "with [Split-Body] I did not have to think too much about the direction of moving the left hand, so I could focus on the [n-back] task".Of the two participants who did not mention a perceived improvement with SplitBody, P11 stated "[I did] a little better with no EMS (. ..) but pretty close [on both conditions] for the computer task", and P3 stated that they "performed better on the movement task without EMS but did better in the [n-back] with EMS."These are perceived scores, not their scores; in fact, P3 performed marginally better with SplitBody (2%) and P11 performed 40% better with SplitBody.
EMS distractions.Overall, seven participants specifically stated they did not find EMS distracting (P1, P2, P4, P5, P6, P7, P12), while five still found it distracting (P3, P8, P9, P10, P11).Of the participants who stated not to find EMS distracting, P2 noted "I wasn't even focused on it (. ..) since I was focused on the [n-back]", while P6 stated "the EMS intensity was just enough that I could put it in the background." On the other hand, three participants (P8, P9, P10) mentioned that the EMS was, at times, too strong and distracting.To this end, P10 stated "I felt like the pulse was strong, and I would forget the letter I was on, but for the most part, it was fine." Trust in EMS' performance.All participants, except P6, stated they trusted the EMS automation to perform the correct pattern.They expressed either forgetting about it, such as P9 stating "I did not think about it at all.I did not even think about if it was doing the wrong pattern", or feeling that it was correct, as P10 stated, "I did not question it (. ..) they seemed correct".P6 stated being skeptical at times, saying, "I was trying to optimize the letter task.There were times I was doubting [the EMS movements]; it feels like at times, I think it wasn't going in the right direction of the sequence, but I was also not focused on it, so it was hard to keep track".

Discussion & study limitations
Summary of findings.Taken altogether, our results (i.e., reduced overall workload and mental-demand) support our main hypothesis (H1, i.e., our SplitBody condition's ability to automate one of the tasks would result in a decreased mental workload).We also observed increased performance & faster response time, supporting our second hypothesis (H2, i.e., reduced mental workload improves task performance).
Study limitations.Our study is not without its limitations.First, we focused on creating a challenging multitask and thus resorted to one that combines muscle movements and cognitive operations (e.g., short-term memory).Yet, this is only one of many possible physical multitasking situations that users encounter, so we advise to be mindful in extrapolating our results to tasks that are fundamentally different.Secondly, the ability of the EMS actuation to assist users with complex movements is limited by the capability of EMS research (e.g., many explore how to make it more precise [2,7,16,31,34,37,46,63,86,91].In fact, our observations lead us to believe that there are per-participant differences in how well participants let the EMS move their hands (e.g., we noticed that three participants tensed up their non-dominant hand as they are not used to the feeling of having their limb move involuntarily, and hence decreasing the quality and precision of each EMS stimulate).This points to an open challenge in EMS research in optimizing how natural these actuated movements feel to the user.

ENVISIONED APPLICATIONS
We illustrate the use of SplitBody in four envisioned applications, in which our system assists by performing repetitive background tasks: (1) writing while cooking; (2) drawing while coloring; (3) soloing on the snare-drum while playing a backbeat; and, (4) playing an instrument while being accompanied by another.
We designed these applications to highlight scenarios that are not meant to be automated by a machine, but instead, where users seek to be physically engaged in the task, either for the sake of creativity (e.g., drawing), learning (e.g., playing music), or pleasure (e.g., cooking).These were chosen to convey how SplitBody can open up a new design space for interactive EMS systems.
Also, these examples were designed by taking into account the precision of existing EMS systems.In fact, the interactions depicted were designed conservatively with respect to the accuracy already possible with EMS.

Split-chef: making caramel while writing an essay
In our first envisioned application, a user multitasks by preparing caramel while writing an essay.Making caramel demands constant stirring and monitoring to prevent burning.Our user taps on their EMS device, activating a pre-programmed stirring motion with SplitBody.This allows them to divert their focus to writing, as shown in Figure 6 (a).Once they feel the sugar is consistently melted, as their sense of proprioception lets them feel the change in viscosity while stirring (even though this hand is EMS-controlled, proprioception is never off [46]), they suspend writing to switch their focus to add butter into the pot, finalizing the mixture, as depicted in Figure 6 (b).Subsequently, they switch back most of their focus to writing as the EMS continues stirring the added butter.Upon achieving the desired caramel consistency, they tap the stimulator to stop SplitBody.While this application is envisioned due to its simplicity (e.g., open-loop-EMS, no-tracking), the user in Figure 6 (c) was indeed successful at cooking/writing with SplitBody while avoiding burning the caramel (this caramel was taken home for their enjoyment).Technical feasibility.The EMS movement used in this application was inspired by Kaul et al. 's [31], which demonstrated that EMS can actuate a user's arm in a circular motion with an average error of 17.8mm-this system's implementation & accuracy would be sufficient to realize our proposed stirring gesture.

Split-draw: enabling synchronous shadow drawing and coloring while sketching
In this envisioned application (inspired by the Split Body artwork of Stelarc [84], to which our system's name is an homage), we explored drawing with SplitBody: (a) shadow-drawing and (b) coloring.While these are envisioned explorations, it is possible to track & actuate a user's drawing with an EMS system similar to Muscle-Plotter [49].First, Figure 7 (a) depicts a user drawing a house under a sun.Because the user drew a sun, SplitBody actuates the user's non-dominant arm to simultaneously draw the shadows cast by the house without needing to shift all of their attention.Second, Figure 7 (b) shows SplitBody coloring inside a shape that the user just finished drawing, allowing the user to move to the next shape, while the coloring process continues as a background task.
Technical feasibility.The EMS used is similar to Muscleplotter's [49], which achieves a drawing accuracy of ±4.07mm-this system's implementation & accuracy would be sufficient to realize our proposed drawing gestures.

Split-drum: learning to drum one limb at a time
In this envisioned application, we depict a novice drummer playing a full drum set without, yet, being able to multitask on each drum kit's part (a hard skill when learning drums, referred to as limb independence [95]).Figure 8 depicts our user choosing to have SplitBody automate the backbone of a funk drumbeat (i.e., EMS plays the bass & hi-hat) while, voluntarily, the user focuses more of their attention on playing the snare at the correct timings.Technical feasibility.The EMS used for drumming is inspired by Ebisu et al.'s [13], which demonstrated EMS' ability to play rhythms with both hands (using more complex beats than ours).Also, EMS lends itself well to timing-based applications, such as the envisioned drumming, due to its fast actuation speed (e.g., 40ms for [29]).

Split-musician: alternating foreground & background musical tasks
Finally, we explored the concept where users alternate between which task is automated with SplitBody and which task is performed voluntarily.Figure 9 (a) shows our user soloing on the synthesizer while letting SplitBody play the drum.Then, as depicted in Figure 9 (b), our user decides to swap these around by pressing a footswitch, which causes SplitBody to, in the background, play simple threenote arpeggios on the synthesizer, while the user redirects more of their attention to playing a more advanced rudiment pattern on the drum.Technical feasibility.The EMS used to move individual fingers in this application is based on Takahashi et al.'s [86], which demonstrated that EMS can actuate all four fingers with an index of independence of 0.62-their approach's accuracy is sufficient to play single keys on the synthesizer as we depict in this application.

DISCUSSION
Safety & Ethics.We believe that any interactive system with the capability of moving the body must be ethically designed by grounding it in the principles of user-agency & safety.This is precisely the case for SplitBody.First, while our explorations were entirely lab-based, in all these situations, our users were given full control of when to activate SplitBody's EMS (e.g., pressing a button while cooking, pressing a footswitch while drawing, etc).In other words, SplitBody does not include automatic triggers that invoke EMS assistance, only user-defined triggers.This mechanism further implements a simple way for users to turn off the EMS assistance.Importantly, SplitBody only actuates a subset of muscles (e.g., forearm & wrist muscles while cooking, wrist & calf muscles when drumming, etc), always leaving most of the user's limbs nonactuated and completely under the user's voluntary control-this allows the user to turn off the assistance when desired.Secondly, as with other interactive systems based on EMS, we believe that fully realizing any of SplitBody's applications outside of a research environment must include features that provide agency to the user, such as: automatically halting any EMS when the user moves against the stimulated movement (e.g., as used in [47]), providing user-defined gestures that immediately suspend the stimulation (e.g., as used in [46]), or only enabling the stimulation in user-defined areas (e.g., as used in [47] or [49]).Moreover, all our experiments followed the established EMS guidelines [36,65,80], were approved by our ethics review board, and conducted with the informed consent of all participants.Conceptual differences to external automation.Unlike devices that automate background tasks using machinery (e.g., robots) that are external to the user, interactive actuation systems (e.g., EMS) act on the user's body.While EMS-actuated movements feel less agentic than one's own voluntary movements [29,30,85], users are still involved as they feel their body moving via their sense of proprioception/touch [46,89].This distinction is key to our concept, which focuses on interactive contexts where utility is not the user's sole objective (e.g., unlike tasks viewed as chores such as vacuuming by hand), and, instead, the user's goals involve not only utility but also body-ownership [12,54].As such, we focused on situations where users want to automate repetitive gestures but also want to be bodily-engaged with the tasks-for the sake of their creativity, learning, pleasure, or even for a sense of ownership over the outcome [12,54,55,85]-rather than letting an external machine perform the background task for them, which offers no sense of involvement due to the "full automation" [5,52].Naturally, we acknowledge there are many scenarios where background automation is beneficial using external devices and where users might feel no desire to be bodily engaged (e.g., vacuuming robots), which were not the focus of our investigation.

CONCLUSIONS & FUTURE WORK
We proposed and evaluated a novel concept (SplitBody) that uses electrical muscle stimulation to assist users in movement tasks that happen in the background while the user is focusing on another task in the foreground.We found that participants assisted with SplitBody were able to perform better on a physically demanding multitask.Inspired by these findings, we envisioned a set of applications to illustrate the design space that SplitBody opens.
Future renditions.While EMS is a highly portable actuator capable of moving the body, other force-feedback actuators (e.g., exoskeletons [8,18,38,50,66], artificial-muscles [17]) may exhibit more precision at the expense of their larger form-factor.As our concept hopes to one day integrate with everyday interactions, it was important for us to choose a small & portable device.That being said, we expect that similar benefits can be found using mechanical devices, and we hope that our work inspires future work in that unexplored direction.
Future integration with supernumerary limbs or VR.We believe that some of the advantages of SplitBody might be integrated with supernumerary-limb interfaces [56,78], such as the decreased mental workload or the ability to alternate between automated/voluntary tasks.Similarly, researchers started to explore how users simultaneously control two VR avatars [35] (also a type of SplitBody, but for input); we believe that our SplitBody might provide useful haptic feedback so that these VR users can synchronize their body pose with their virtual avatars prior to initiating control.

ACKNOWLEDGMENTS
We would like to thank Prof. Peggy Mason for insightful discussions on this topic.This work was supported by NSF grant 2047189 and 2212352.Any opinions, findings, conclusions, or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of any funding agencies.

SPECIAL ACKNOWLEDGMENTS
We would like to acknowledge that the ACM CHI 2024 conference has decided to take place in Hawaii, offering extremely limited options for remote participation.While some authors have chosen to present their work in person or bring its physical prototype to the CHI community, they have done so with compassion towards Native Hawaiians and with a heavy heart.We urge that the ACM's process for conference selection should more carefully consider the impact our conferences have on local communities.We especially want to acknowledge Prof. Josiah Hester, a Kanaka Maoli Professor in Computing, who organized comprehensive resources that educated us about the negative impacts of over-tourism and climate degradation in Hawaii2 .Finally, the decision of some authors to participate in this conference does not represent the views of other members of our lab, who have chosen not to engage with this edition of the CHI conference due to its impact on Native Hawaiians.

Figure 2 :
Figure 2: (a) In the baseline condition, the participants are performing a movement task and a n-back task by themselves.(b) In the SplitBody condition, the participants are performing the same tasks, but the movement task is being automated by EMS (lightning icon).In both conditions, the movement task (c) involves repeating the following arm gesture sequence: up, down, left, right, down, up, right, and left.

Figure 3 :
Figure 3: (a) Score with SplitBody and baseline, including a breakdown for the (b) movement (gesture) and (c) cognitive (n-back) tasks.

Figure 4 :
Figure 4: The NASA Task Load Index score for both conditions.

Figure 6 :
Figure 6: (a) A user is multi-tasking by making caramel with SplitBody on their left arm (lightning icon) while they are writing an essay with the other arm (solid green arrow depicts their main attention).By feeling the consistency of the melted sugar in the pot, (b) they switch their attention back to the cooking caramel (solid green arrow depicts the switch of their main focus) and butter to the mix while their left arm is still automated by EMS (lightning marker).Finally, (c) they stop SplitBody by taping on the stimulator.

Figure 7 :
Figure 7: A user drawing (a) simple shadows and (b) coloring with SplitBody on their non-dominant hand (lightning icon) while continuing to draw with their dominant hand.

Figure 8 :
Figure 8: A novice drummer only focusing on playing the snare drum while SplitBody automates the hi-hat and bass (lightning icon).

Figure 9 :
Figure 9: A user switching between which task is automated with SplitBody (lightning icon) and which task is performed voluntarily.