The most common haptic effects one encounters in consumer devices are produced by mechanical actuators. The primary function of such devices is to deform the skin to produce various tactile cues, such as, buzz, rumble, and click. Although the sensations produced using these methods are compelling and reproducible, the gamut is highly limited. Ideally, haptic feedback for human-machine interfaces-for example, for education, entertainment, healthcare, and medical devices-should be able to produce a much wider range of tactile cues (1)(2)(3)(4). Indeed, haptics have become indispensable for consumer devices, such as, "vibration mode" on a phone. Similarly, patient engagement and usability devices envisioned for stretchable, epidermal sensors could be substantially deepened with this type of unobtrusive cue. Moreover, use cases for epidermal appliances in physical therapy, robot-assisted surgery, and remote operation could be expanded if they could provide haptic feedback.
All tactile sensations which are ultimately perceived by the brain originate as action potentials in afferents of the peripheral nervous system. Thus, it is in principle possible to generate any tactile percept by activating the relevant set of mechanosensory neurons. Transcutaneous electrical stimulation of the mechanoreceptors using electrodes pressed to the skin has thus held promise for many decades, for applications such as haptics, pain relief, and neuroprostheses (5). This electrotactile effect is thus a form of "sensory substitution," in which electrical signals can be perceived as mechanical forces (6). However, electrotactile stimulation as a means of generating haptic effects has been criticized for many reasons (7). It is non-selective, meaning that it produces a range of sensations which are inconsistent between users. This inconsistency owes to differences in skin morphology, unequal hydration, and other parameters which affect contact impedance with the electrodes (8). Moreover, high currents and voltages required when using conventional materials (such as metals) and device layouts (like hard planar substrates) also lead to unwanted effects, such as galvanic reactions at the skin surface, pain, and temporary desensitization (9). Also, the area of stimulation tends to be large, and electrical signals also have the potential to stimulate the muscles as well as the mechanosensors (10).
Nevertheless, the toolkits of materials chemistry, microfabrication engineering, and signal detection theory applied to cognitive science have the potential to increase the realism, safety, and reliability of electrotactile effects (11). Such development would greatly enable haptic effects in epidermal circuits, where devices in wearable form factors would otherwise have to be made thicker to accommodate components with moving parts (12)(13)(14)(15). In this article, we use an integrated materials design strategy that combines a purpose-synthesized, stretchable, conductive block copolymer, concentric electrodes in a stretchable layout, and a psychophysical design strategy for consistent stimulation of mechanical sensations in human participants (8,16,17).
Using these tools and a large data set, we were able to show extraordinarily low stimulation currents (≥6 µA), improved spatial localization, greater acuity by the participant, and the ability to toggle between sensations characterized as pressure versus vibration by modifying the frequency of the signal.
Our strategy is illustrated in Fig. 1. The material (Fig. 1A) is a conductive block copolymer based on the well-known polyelectrolyte complex, poly (3,4ethylenedioxythiophene):poly(styrenesulfonate) (PEDOT:PSS). To circumvent the stiffness and brittleness of pristine PEDOT:PSS, which does not make conformal coverage with the uneven surface of the skin, we used a block copolymerization strategy described recently by our group (18,19). The goal was to make a low-impedance conductive polymer with a high degree of mechanical deformability, and which can be patterned easily into conductive traces. Briefly, a flexible bottlebrush segment based on poly(poly(ethylene glycol) methyl ether acrylate) (PPEGMEA) was covalently bound to the PSS chain using the reversible addition-fragmentation transfer (RAFT) process, a type of quasi-living controlled radical polymerization. The resulting polymer was termed PEDOT:PSS(1)-b-PPEGMEA(6) ("Block-6"), where the subscripts refer to the mass ratios of the two segments (figs. S1 to S5). Inclusion of the large, flexible units resulted in a much lower elastic modulus than PEDOT:PSS (~10 MPa compared to ~380 MPa) (18). A molecular dynamics simulation of a single polymer chain with associated PEDOT (Fig. 1B) revealed the flexible PPEGMEA chains surrounding the stiffer PEDOT:PSS core, reminiscent of common thermoplastic block copolymer elastomers, such as poly(styrene-butadiene-styrene) (SBS). The low elastic modulus of Block-6 allowed for notably greater conformal contact on the surface of the skin (Fig. 1C and fig. S6). Poor contact leads to increased impedance, larger currents, and greater chance of pain upon stimulation (20,21).
Using laser ablation, we fabricated a design consisting of concentric electrodes in a serpentine layout (Figs. 1D and 1E). Although the addition of serpentine microstructures may not be groundbreaking, it's important to acknowledge that integrating them into our intrinsically stretchable material offers notable advantages. Our material inherently stretched up to approximately 10% elastically, and the incorporation of serpentine microstructures extended its stretchability, making it suitable for use as a skin interface. The details of fabrication are given in (figs. S7-10). Briefly, the Block-6 electrodes were deposited atop a thin film of poly(methyl methacrylate) (PMMA) on a stretchable substrate, polydimethylsiloxane (PDMS). We chose to test for tactile effects on the forearm because it represented a greater challenge for controlled tactile effects compared to the fingers, which have a lower stimulation threshold and higher afferent density (22)(23)(24).
Recent studies indicate that a range of current densities (0.1 -10 mA cm -2 ) can stimulate sensory fibers residing at the epidermis-dermis junction (7). In addition, the distribution of the low current density should be localized, to activate nearby mechanoreceptors (and not muscles). Thus, to stimulate locally, a concentric design was adopted for the ventral forearm (Fig. 1E and F and figs. S11 and S12). Finite element analysis (FEA) simulation showed that the current density distribution was localized across the skin layers as opposed to monopolar designs under similar area (~10 cm 2 ) and current source (10 µA) (Fig. 1G and figs. S13). The electrode dimensions were optimized to fit the forearm and the current applied is the average detection threshold from the participants (~10 µA) (fig. S13). The electronic properties utilized for current density simulations are shown in fig. S14 and Table S1. Importantly, we assessed the effectiveness of current density distribution on an embedded nerve fiber in the epidermis layer by implementing the Hodgkin-Huxley model, which describes action potential propagation (Table S2). Schematics illustrating the nerve fiber location and embedment in the dermis are shown in fig. S15 and fig. S16. The model suggests that the action potential only propagates along the fiber when a concentric design is placed on the skin, as it remains untriggered in a conventional electrode due to the non-localized current spread (Fig. 1H and fig. S17). Furthermore, we conducted an evaluation of the dual electrode with a comparable total contact size (505 mm 2 ) to serpentine concentric (509 mm 2 ) and gap distances of 20 mm and 9 mm, as depicted in fig S .13. This assessment revealed an increase in current density compared to the larger dual electrode, and it indeed triggered an action potential at a 9 mm gap distance (fig S.18). However, the gap distance is usually 20 -50 mm for dual electrodes (25). It is essential to note that existing reports generally favor the use of larger electrodes to minimize the likelihood of inducing painful sensations (26)Consequently, the geometric shape employed influences the magnitude of the current stimulation and the subsequent sensation.
The electrochemical characteristics of the Block-6 electrode influence how charge is transported across the electrode/skin junction. For example, the capacitance of the electric double layer (EDL) within a conductive polymer determines the amount of charge that can be injected (27,28). PEDOT:PSS exhibited capacitive behavior due to the charged sulfonate groups; these charges permit mixed (ionic and electronic) conductivity (Fig. 2A and fig. S19A) (29). The cyclic voltammetry (CV) of Block-6 showed capacitive behavior by the rectangular shape hysteresis in the voltammogram, and the linear dependence of the current with the scan rate (fig. S19B) (30). In addition, the electronic function of the electrode is dependent on both the electrical and ionic conductivity of Block-6. Although the electrical conductivity of Block-6 is lower than that of pristine PEDOT:PSS, the PPEGMEA chains promote ionic transport, and thus resulted in a greater EDL capacitance, indicated by a greater slope when measured from CV (fig. S19B) (29,30). The CV additionally exhibited a larger area compared to pristine PEDOT:PSS, suggesting a larger charge storage capacity of Block-6 (Fig. 2B and fig. S20). Finally, the excursion potential of Block-6 and pristine PEDOT:PSS were measured with a charge injection of 2.08 mA mm -2 , and both electrodes showed comparable stability in terms of voltage build-up (Fig. 2C). This is important for safety consideration during charge injection (31). However, increasing the injection current resulted in delamination of PEDOT:PSS (fig. S21), whereas the injection current of the Block-6 electrode could be increased to 36 mA (75 mA cm -2 ) while remaining intact.A critical consideration for wearable electrodes is compatibility with elastic human skin (32). Delamination due to mismatch in modulus can result in increased impedance, increased voltage required for simulation, and greater chance of pain, galvanic reactions, and other unwanted effects. The viscoelasticity of Block-6 permits the local conformability to the uneven surface of the skin, and also lateral stretching due to global deformation of the skin (Fig. 2D, fig. S22). Microscopic conformability in particular is unfeasible for high-modulus metallic films (33). Here, the concentric Block-6 electrodes were designed to have serpentine structures to accommodate increased elastic strain and reduced changes in interfacial impedance upon stretching (Fig. 2E). In addition, impedance measurements on human skin were lower for Block-6 across all frequency ranges (fig. S23). Mechanical simulations suggested that the elasticity of the freestanding Block-6 electrode to be ~44% (fig. S24 and fig. S25). Although the elasticity of the assembled device decreased to ~19% due to the bonded contacts on the elastomeric substrate (fig. S26), this still remains a ~3× improvement compared to recent reports of metallic interconnects on elastomers (34). Likewise, strain-resistance measurements indicated that the Block-6 electrodes exhibited negligible piezoresistivity-meaning no increase in resistance up to 30% for at least 500 cycles (fig. S27). Also, we characterized the contact resistance between silver for Block-6 (0.217 Ohm•cm 2 ), which is comparable to pristine PEDOT: PSS (0.237 Ohm•cm 2 ) (fig. S28). Additionally, simulations suggest that torsion of the electrodes up to 180º results in von-Mises stresses that are less than half of the yield strength (Fig. 2G). Finally, the intrinsic stretchability of the polymer coupled with the serpentine design allows the device to achieve strains of 100% before failure (Fig. 2F). The PPEGMEA chains render a lower loss and storage modulus (Fig. 2H and fig. S29), with higher viscoelasticity than that of pristine PEDOT:PSS (fig. S29 and S30). When compared to physical blends involving PEDOT (Fig. 2I), the covalently engineered Block-6 possesses the lowest modulus, indicating it is closest to that of human skin.
Both electrical and mechanical stimuli can activate nerve fibers that evoke the sense of touch. Mechanical stimulus results in the deformation of mechanoreceptors, whereas electrical signals produce a potential gradient across nerve fibers, resulting in depolarization and repolarization events across the membrane of receptor cells. As a result, these stimuli trigger action potentials that ascend towards the central nervous system (CNS) in the same biological pathway (Fig. 3A) (35,36). For mechanical stimulus, the type and intensity of the sensation triggered by activation of mechanoreceptors is a function of displacement and frequency. For example, afferents terminating in the Merkel cell-neurite complexes (slow-adapting type I) are responsible for the sensation of sustained pressure, whereas rapid-adapting mechanoreceptors (RAMs) are sensitive to vibrotactile stimulation (37,38). Aβ-fiber RAMs in the hairy skin are tuned to very similar frequencies as those terminating in Meissner corpuscles (rapid-adapting type I) deliver low vibration sensations and are believed to make up part of the sensation of fine textures.
To relate the parameters of electrical stimulation to the sensation perceived, we performed psychophysical experiments on a group of participants (Fig. 3B). Safety protocols were assessed before experimenting with stimulation (fig. S31 and S32). We designed a set of psychophysical tests that will characterize key elements in electrotactile stimulation, including pain and detection thresholds, perception quality, spatial acuity, and contrast discrimination. Detailed psychophysical tasks and procedures are described in fig. S33 to S35. Furthermore, we have demonstrated that materials like pristine PEDOT: PSS (known for rigidity) and copper (characterized by high impedance) are impractical for electrotactile applications (fig. S36 and fig. S37). The psychophysical evaluations began with pain threshold measurement by applying biphasic current stimulation (10 -250 µA). Measuring pain thresholds enables us to know if a participant might feel noxious on subsequent tasks. Fortunately, the highest intended stimulation current for the detection threshold experiment was ~4× lower than the lowest pain threshold of any participant (Fig. 3C). To quantify the detection threshold associated with the haptic sensations invoked, we used a 2-alternative forced choice (2AFC) design to measure the accuracy with which participants could identify which of two electrotactile devices had been stimulated. Even at the lowest stimulation current of 6 µA, participants' accuracy was notable, reaching 84%, which was significantly higher than would be expected by chance (p < 0.005; see Statistical Analysis), and higher than the 75% accuracy criterion classically used to define the sensory detection threshold (Fig. 3D and fig. S38) (39). To our knowledge, ≤ 6 µA is among the lowest sensory detection threshold reported for an epidermal electrotactile device.
Idiosyncrasies of the participants, such as differences in skin morphology, hydration, and concentration-can lead to individual differences in sensation (10). In subsequent experiments, we accounted for these individual differences by applying a current amplitude equal to (1T), twice (2T) or three times (3T) the detection threshold level for each participant (rather than applying fixed currents across participants). We hypothesized a gamut of percepts by normalizing the stimulation frequency. Across all combinations of frequency and current of electrical stimulation, participants predominantly experienced the sensation of either pressure or vibration. In fewer than 5% of trials, a different sensation was reported (typically itch, possibly due to the activation of pruriceptors (40). Overall, the participants were significantly more likely to experience vibration at higher frequencies of stimulation, and pressure at lower frequencies of stimulation (p < 0.001) (Fig. 4A and fig. S39) (41). On average, increasing the stimulation frequency by 70% doubled the odds of experiencing vibration versus pressure. Our model predicts that the "crossover" point at which participants first begin to report vibration more often than pressure is ~6 Hz (Fig. 4B). Interestingly, the frequency of the crossover point happens to correspond to the frequency at which Meissner corpuscles (sensitive to low-frequency vibrations) become more sensitive to vibration than Merkel cell-neurite complexes (which typically respond to sustained pressure, but are also sensitive to very low-frequency vibrations) (11,42). Aβ-fiber RAMs in hairy skin are sensitive to frequencies between 10 and 50 Hz, similar to the range that activates fibers terminated in Meissner corpuscles found in glabrous skin. This observation suggests that the change in conscious sensation from pressure to vibration might reflect an equivalence (in the elicited patterns of neuronal activation) between electric stimulation and mechanical stimulation at a given frequency (Fig. 4C).
Although the type of sensation (pressure versus vibration) was strongly dependent on the frequency, it was not significantly affected by the current (p > 0.05). Instead, changes in stimulation current affected the intensity of the sensation perceived, with stronger currents corresponding to a more intense sensation (for instance, doubling the stimulation current increased the perceived intensity by approximately 0.25 on the 1 -5 intensity scale, fig. S40). Importantly, as predicted by the pain thresholding, all sensations were reported to be between "very comfortable" and "neutral" except for a single trial for a single participant (who rated the 10 Hz stimulation at detection threshold as "uncomfortable"). Furthermore, almost all sensations (93%) were experienced as localized rather than referred (5%) or radiating (1%). The probability of experiencing a sensation as localized did not significantly depend on frequency or current (both p > 0.05; Fig. 4D). Moreover, we have assembled a comprehensive table for comparing our work to the existing literature (Table S3).
Finally, the types of sensation, spatial acuity, and contrast discrimination achievable by electrotactile means will depend on the type and density of nerve endings found on various regions of the body. Previous studies suggest that the spatial acuity for the forearm by electrical stimulus (a type of two-point discrimination test) is 9 -40 mm (7,43,44). Using our materials, we varied the inter-electrode distance between two electrodes from 2 -10 cm (fig. S41). The smallest distance between electrodes tested was constrained by the outer terminal of the concentric electrode, which had a radius of ~1 cm (fig. S11 and S12). Even at the smallest possible centerto-center distance, 2 cm, participants could identify whether stimulation came from one versus both electrodes with 95% accuracy (p < 0.001; fig. S42). Overall, the identification of spatial acuity was significantly higher than chance (p < 0.001). There was not a significant relationship between inter-electrode distance and acuity accuracy (p = 0.53). However, increasing the difference in current between the electrodes at a fixed position (meaning the just noticeable difference) corresponded to better accuracy (p < 0.001) (Fig. 4E). On average, increasing the "contrast" in signal by 13% doubled the odds of identifying which site had higher current.
We conducted a comparative analysis of our Block-6 psychophysical results with silver/silver chloride (Ag/AgCl) commercially available electrodes with identical experimental settings (fig. S43A). Hence, it is possible to directly compare the electrodes at different amounts of current delivery. The detection performance at 6 µA was far lower than the accuracy criterion (fig. S43B). Therefore, we increased the current amplitude up to 80 µA for further investigation of the detection threshold (fig. S43C). We found that Block-6 outperforms the commercial electrode and the minimum current for above chance performance on the commercial electrode is ~25 µA (~3x higher current than with Block-6). Also, the commercial electrode crosses the traditional psychophysics threshold definition of 75% accuracy at ~40 µA (~5x higher current than with ours). Moreover, the frequency modulation resulted in more thermal and itching sensations (fig. S43D). The toggling effect between pressure and vibration was imperceptible with this type of electrode. Lastly, the dual electrode experiences more current spread and thus the response proportion was not localized at all frequencies (fig. S43E).
The generation of haptic effects by direct electrical stimulation of afferent fibers has always held promise as a means of sensory substitution. Here, we have shown how an integrated strategy that includes polymer engineering, device layout, and psychophysical thresholding based on signal detection theory can be used to overcome the occurrence of pain, desensitization, lack of realism, and participant-to-participant variation which has previously hindered this technique. These methods allow delivery of ultralow stimulation currents with high spatial precision. We also demonstrated the ability to "toggle" between sensations categorized by participants as pressure and those categorized by participants as vibration, by adjusting the frequency of electrical stimulation. Moreover, the crossover point in electrical frequency at which perception of vibration is more likely than perception of pressure is similar to the mechanical frequency at which Aβ-fiber RAMs, a subset of low-threshold mechanoreceptor activated by low-frequency vibrations, become more sensitive than Merkel afferents. This observation suggests an equivalency between electrical and mechanical stimulation. The ability to bypass the mechanically sensitive apparatus and activate the appropriate neurons transcutaneously could be valuable not only in haptics but also in neural prostheses.
We note that several observations in this study may open new avenues for research. For example, our results hinted at the possibility of a type of "tactile contrast": increased discrimination when different electrical signals were supplied by adjacent electrodes. This capability could be highly valuable in the development of tactile displays, human-machine interfaces, accessibility devices, or new forms of artwork, especially for the blind. Furthermore, the results from our cohort in which participants categorized (a small number of) electrotactile stimuli as "itch" or "heat" (in addition to the dominant categories of pressure and vibration) suggest a broader gamut of tactile percepts to the electrotactile effect than may have been previously known. Such results also hint at the ability to achieve multimodal stimulation given a more complex waveform or juxtaposition of electrodes providing different signals.
Finally, we note that the technical sophistication of our electrode setup is not the best that can possibly be achieved. Excellent work of many investigators in the field of epidermal electronics have demonstrated stretchable layouts with greater resolution than those reported here. More sophisticated layouts with multiplexed outputs and tighter resolution-combined with the materials and psychophysical calibration reported here-may still lead to new capabilities. Fig. 1. Intrinsically stretchable organic and soft electrode for electrotactile applications. (A) Schematic illustration of intrinsically stretchable PEDOT derivative known as Block-6. (B) Atomistic molecular dynamics simulation of Block-6. (C) Schematic illustration of (i) Block-6 and (ii) pristine PEDOT:PSS conformability on corrugated skin (iii) High-resolution optical images displaying conformability of the electrode on porcine skin at 200X magnification . Scale bar, 100 μm. (D and E) Photographic images of conformal and stretchable concentric electrodes on human skin for electrotactile applications, including the fingertip (D) and the forearm (E). Scale bars of 1 mm and 1 cm, respectively. (F) Schematic illustration showing the comparison of current flow in serpentine concentric and dual electrode designs. (G) Simulation results of current density distribution in dual and serpentine concentric electrodes. Large electrode has a contact area of 2217 mm 2 . Dual and serpentine electrodes have comparable contact areas of 505 mm 2 and 509 mm 2 , respectively. (H) Hodgkin-Huxley model shows nerve actional potential triggered in serpentine concentric electrode design as a result of localized current density. means of GLMM fixed effect (black line) and 95% confidence interval on fixed effect (shaded grey ribbon) (n = 400 trials).
Number-average molecular weight (Mn), weight-average molecular weight (Mw), and dispersity (D̵ ) were determined using an Agilent Technologies 1260 Infinity II LC system. The mobile phase was 30% methanol and 70% 0.2 M sodium nitrate (NaNO3) and 0.01 M monosodium phosphate (NaH2PO4) in water at pH 7 (adjusted with concentrated sodium hydroxide, NaOH) at 40 °C at 1 mL min -1 . The PL aquagel-OH Mixed-B column was used, calibrated against narrow dispersity PSS standards (purchased from Polymer Standards Service). 1 H nuclear magnetic resonance (NMR) spectra were acquired in deuterium oxide D2O at room temperature on a Bruker AVANCE III 600 MHz NMR spectrometer fitted with a 1.7 mm triple resonance probe with the z-gradient.
Sodium 4-styrenesulfonate (NaSS), 4,4′-azobis(4-cyanovaleric acid) (ACVA), azobis(isobutyronitrile) (AIBN), PEGMEA (Mn = 480 g mol -1 ), 4-cyano-4-(phenylcarbonothioylthio)pentanoic acid (the reversible addition fragmentation transfer (RAFT) chain transfer agent), and ethylenedioxythiophene (EDOT) were purchased from Sigma-Aldrich and used without further purification. Distilled water filtered using a Milli-Q purification system was used throughout.
RAFT Polymerization of PSS(1)-b-PPEGMEA( 6) block copolymer PSS(1)-b-PPEGMEA(6) was synthesized as previously described (18). Briefly, PSS macro-RAFT was synthesized by RAFT polymerization of sodium styrene sulfonate (NaSS) monomers. The PEDOT:PSS(1)-b-PPEGMEA( 6) was synthesized as previously reported (18). Briefly, PSS(1)-b-PPEGMEA(X) was dissolved in MiliQ water, at a concentration of 75 mg mL-1 and was acidified by Dowex Marathon C hydrogen form for 18 h. The polymer solution was filtered over a 10 μm filter. Sodium persulfate (1150 mg, 4.83 mmol) and iron trichloride (100 wt % in water, 0.175 mL) were added to the filtrates of 4875 mg of PSS(1)-b-PPEGMEA( 6), 65 mL. The solution was vigorously stirred before the addition of EDOT (500 mg, 3.51 mmol). MiliQ water was added in small portions to prevent gelation. The final volume was 180 mL. The reaction was left to react for 24 hours at room temperature. PEDOT:[PSS(1)-b-PPEGMEA( 6)] was purified by stirring over acidic resin (Dowex Marathon C hydrogen form, 17 g) and basic resin (Lewatit MP-62 free base, 11 g) for 6 h, followed by a 10 μm filtration.
A custom mold made from Clear V4 resin was 3D-printed (Form 3+) with a thickness of 0.5 mm for drop casting. The mold was sonicated with isopropyl alcohol (IPA), acetone, and deionized (DI) water, sequentially and then dried with nitrogen (N2) gas. Mold release (Ease Release 200, Mann Release Technologies) was sprayed on the 3D-printed mold and was let to dry at room temperature. Block-6 ink was vigorously stirred and degassed for at least an hour each prior to drop casting. A disposable plastic syringe (1 mL, Henke-Ject) was used to disperse 6.5 mL of the solution. The film was dried on a hot plate covered with a Pyrex tray at 65 °C for 1 hour and 30 min. The film thickness was measured with a profilometer (~20 µm, Bruker Dektak XT stylus).
The film was peeled using tweezers and stored in a plastic storage container.
Participants' recruitment and safety protocol assessment
The experimental protocol was approved by the Internal Review Board at the University of California San Diego Human Research Protections Program. Ten healthy volunteers were recruited from the University of California San Diego participated in this study. This sample size fits a pilot study size (14). All steps were taken to ensure a diverse set of participants. The participants had no visible skin cuts or infections in the application area. The participants were blinded for the input of stimulation. Inclusion criteria comprised individuals from a healthy demographic, aged between 18 and 40 years. The age range of 18 to 40 was selected due to the lower likelihood of chronic diseases in this age group. Exclusion criteria included: pregnancy, absence of skin injuries on or near the stimulation sites, rubber or elastomer allergies, highly sensitive skin, presence of pacemakers or other electrical or metal implants in the body, epilepsy or heart conditions, self-reported cancer or neurodegenerative diseases, and allergic reactions to topical moisturizers.We evaluated the safety of our electrodes during stimulation by thermal infrared imaging to detect any joule heating effects (Fig. S31). In addition, we measured the source voltage required to stimulate currents on the finger and the forearm in contrast to recent reports (fig. S32). After assessing safety protocols, our ten human participants were subjected to experiments with the haptic device on the ventral forearm. Biphasic current stimulation is advisable to minimize pain perception and avoid half-cell reactions (45). Quantifying pain thresholds enables us to know if a participant feels noxious for succeeding tasks.
Before placing the electrode on the skin, we wanted to ensure skin condition was similar as possible between different participants because it has a strong influence on the variables, and thus affects the sensation and comfort of stimulation. To remove the insulating layer of the stratum corneum that bears a high impedance of 100 kΩ, the skin was scrapped gently with 3M Red Dot TM Trace Prep and with 70% alcohol prep pad (11). In addition, since the conductive path through the skin is not uniform at the microscopic level due to minute epithelial breaks, we used Q-tips to minimally wet the surface with MiliQ water before placing the electrode (20,46).
Stimulation was performed with constant current. Constant current is usually used since sweat can build up (20,47). A symmetric, biphasic, square waveform with 50% duty cycle was used. The square waveform is preferred due to its ease of implementation and interpretation in electrotactile feedback applications. In addition, they also resulted in a faster depolarization of the nerves axon, which promotes effective stimulation (43). Continuous Biphasic was used since it can prevent skin irritation caused by the transfer of ions within axon membranes and it can also provide a wider range of stimulation parameters and more effectively elicit diverse tactile sensations in lower currents it generates a better spread of current compared to monophasic stimulation (48,49). Modulation of one or more parameters, such as amplitude or frequency can render different sensations. In general, the pulse width and amplitude may modulate the perceived intensity, whereas the frequency may modulate the perceived sensation (50,51). For the evaluation of elicited sensations, the frequency below 100 Hz was considered because earlier studies found that low frequency was the most useful range for sensory communication (52). The sensation elicited at 1 Hz and 10 Hz, represented a lower frequency, The sensation elicited at 50 Hz represented a mid-frequency, and the sensation elicited at 100 Hz, represented a higher frequency. The stimulus amplitudes were standardized across participants by using one, two, and three times the detection threshold (DT) of each participant. The stimulation was applied via either the surface electrode to the forearm skin, or to the forehead. A commercial waveform generator and amplifier (BioPac STMISOLA) (fig. S34) were used to generate the studied stimuli. The electrode placement and stimulation setup are illustrated in fig. S33 to S34.
One pair of electrodes on the stimulation site. Participants rated the sensation of pain on a scale of 1 (not painful at all) to 5 (uncomfortably painful). A scale of 1-5 for pain detection was used since there is no uniform definition of pain (20). We stimulated the user's skin using 10 µA and increased the current amplitude in increments of 10 µA every two seconds, until 250 µA had been reached or until the participant reported an uncomfortably painful sensation (5/5 on the scale). The frequency was set to 1 Hz. This process was repeated three times, with a two-minute break between each trial. We set an a priori exclusion criterion that any participant who reported an uncomfortably painful (5/5 on the scale) sensation in the stimulation range 10 -50 µA would be excluded from participating in the remainder of the experiment, no participants did so.
We measured the detection threshold using a 2-alternative forced choice (2AFC) detection task.
Two pairs of electrodes were applied on the stimulation site (either the forehead or forearm). On each trial, only one of the two electrodes was stimulated, and participants were asked on which site they thought they had been stimulated (even if they were unsure, they were required to guess). There was a total of 128 trials per participant: eight trials for each combination of current (8 levels: 6, 8, 10, 12, 14, 16, 18, 20 µA) and stimulation site (2 levels) at a frequency of 1 Hz. The order of trials was randomized within each participant. We chose 16 trials per current level because it conveniently allowed us to operationalize "above detection threshold" using both the classical psychophysical definition (the current at which accuracy exceeded 75%; specifically 13 out of16 trials correct) and the statistical definition (the current at which accuracy was significantly higher than chance on a binomial test; also 13 out of 16 trials correct).
We measured the spatial acuity (two-point discrimination) of the electrotactile current using a 2AFC discrimination task. Two electrode pairs were placed on the ventral side of the forearm, one near the medial epicondyle and the other near the wrist. The center-to-center distance between the inner electrode of the electrode pairs systematically varied from 2 -10 cm in increments of 2 cm.
Stimulation current was set to twice the participant's detection threshold at a frequency of 1 Hz. On each trial, stimulation was delivered to either one or both electrode pairs, and participants were asked whether they thought they had been stimulated in one versus two places (even if they were unsure, they were required to guess). There was a total of 30 trials per participant: three trials for each level of distance (5 levels: 2, 4, 6, 8, and 10 cm) and stimulation type (2 levels: one versus both positions stimulated). There were five blocks of six trials each: one block for each level of stimulus distance. The order of blocks was randomized within each participant, and the order of trials was randomized within each block.
We measured the perceptual quality of stimulation using a self-report questionnaire. One pair of electrodes was placed on the mid-point of the participant's forearm. We systematically varied the stimulation current and frequency to test perception under a wide range of stimulation parameters as follows: 1x detection threshold, 1 Hz, 1x detection threshold, 10 Hz, 2x detection threshold, 1 Hz, 2x detection threshold, 10 Hz, 3x detection threshold, 1 Hz, 3x detection threshold, 50 Hz, and3x detection threshold, 100 Hz.
There was a total of 21 trials per participant: three trials for each combination of stimulation parameters (7 levels; see above). The order of presentation was randomized within each participant. On each trial, participants were asked to answer four questions regarding the sensation they felt, the comfort and intensity of stimulation, and the location of the sensation. The questions are described in detail in the Supplementary Information in the section "Evaluation of the perceptual quality of sensation".
We measured sensitivity to differences in current strength using a 2AFC task. Two pairs of stimulating electrodes were placed on the forearm by applying the detection limit current of the participants. On each trial, a different amount of current was delivered to each electrode. One electrode was set to the participant's detection threshold, and the other was set to one of five different current amplitudes (either 1, 2, 3, 4, or 5 µA higher than the participant's detection threshold). For example, if a participant had a detection threshold of 10 µA, on a given trial one electrode would deliver 10 µA and the other would deliver either 11, 12, 13, 14, or 15 µA. On each trial, the electrode delivering the higher current was varied, and the participant was asked at which electrode they thought the sensation was more intense (even if they were unsure, they were required to guess). There was a total of 40 trials per participant: four for each combination of current (5 levels) and location of higher current (2 levels). The order of trials was randomized within participants.
In general, psychophysical experimental data was fit using mixed effect regression frameworks, generalized linear mixed model (GLMMs; either linear or binomial with logit link function, depending on the dependent variable), which are more powerful than alternatives like t-tests on dprime because GLMMs account for participants' performance on each trial rather than averaging across trials (53). For each experiment, we fit a GLMM with trial outcome as the dependent variable and subject as random effect; the independent variables were experiment-dependent and are described in detail in each subsection below. GLMMs were fit using the lme4 package in R (54), and participant performance at specific levels of the independent variables was assessed using estimated marginal means, calculated using the estimated marginal means (emmeans) package in R (55). All p-values reported in the manuscript are from Wald tests on either regression parameters or estimated marginal means (56).
To quantify the effect of stimulation current on detection accuracy, data from the 2-alternative forced choice (2AFC) detection task were fit using a logistic mixed effect regression (GLMM) with fixed effect of log stimulation current, random effect of participant, and dependent variable of trial outcome (correct/incorrect). Overall, participants were significantly more accurate when stimulation current was higher (GLMM fixed effect 𝛽 𝑙𝑜𝑔(𝐼) = 1.79, 𝑆𝐸 𝑙𝑜𝑔(𝛽 𝐼 ) = 0.25, 𝑊𝑎𝑙𝑑 𝑍 = 7.20, 𝑝 < 0.001); on average, each 60% increase in stimulation current doubled the odds of detecting the stimulus. Nevertheless, even at the lowest stimulation current used in the study, participants' performance was much higher than chance (estimated marginal mean, 𝑃(𝑐𝑜𝑟𝑟𝑒𝑐𝑡) 𝐼=6𝑚𝐴 = 0.84, 95% 𝐶𝐼 [0.66,0.94]), and also higher than the 75% accuracy criterion classically used to define the sensory detection threshold (57). Indeed, we conducted Bonferronicorrected Wald Z tests (𝛼 = 0.006) of the estimated marginal means at all current levels, and participants' accuracy was significantly greater than chance at all eight current levels (all 𝑝 < 0.001).
To quantify participants' two-point discrimination, data from the 2AFC two-point discrimination task were fit using a logistic mixed effect regression (GLMM) with fixed effect of electrode distance, random effect of participant, and dependent variable of trial outcome (correct/incorrect). Overall, two-point discrimination was significantly higher than chance (GLMM estimated marginal mean, 𝑃(𝑐𝑜𝑟𝑟𝑒𝑐𝑡) = 0.96, 𝑆𝐸 = 0.01, 𝑊𝑎𝑙𝑑 𝑍 = 10.68, 𝑝 < 0.001). There was not a significant relationship between electrode distance and two-point discrimination (GLMM fixed effect 𝛽 𝑑 = 0.07, 𝑆𝐸 𝛽 𝑑 = 0.11, 𝑊𝑎𝑙𝑑 𝑍 = 0.62, 𝑝 = 0.53). This null effect of electrode distance is plausibly a ceiling effect due to the physical constraint of electrode spacing: even at the smallest distance possible for our setup (2 cm), accuracy was 93% (estimated marginal mean; 95% 𝐶𝐼 [0.88,0.98]), and for 7 of the 10 participants accuracy was 100%. Note that the GLMM model yielded a singular fit with near-zero variance in the random effect (likely because so many participants had near perfect performance), so to verify that this did not influence our interpretation we reran the analysis as a logistic regression with independent variable of separation distance and a dependent variable of trial outcome (such as dropping the random effect from the model). Doing so did not change the result; indeed, the coefficients are the same as reported above (when rounded to two significant digits).
To quantify two-point discrimination at specific distances (as reported in the main manuscript), we conducted Bonferroni-corrected Wald Z tests (𝛼 = 0.01) of the estimated marginal means. Participants' accuracy was significantly greater than chance at all five distances (all 𝑝 < 0.001).
Qualitative sensation was evaluatedacross all combinations of stimulation frequency and current in the study. Participants predominantly reported the sensation of either pressure or vibration; in fewer than 5% of trials, a different sensation was reported (itchiness or warmth). We classified these trials as outliers and removed from further analysis. To quantify the relationship between stimulation frequency, stimulation current, and reported sensation (vibration versus pressure), data were fit using a logistic mixed effect regression (GLMM) with fixed effects of (log-transformed) stimulation frequency, (log-transformed) stimulation current, and their interaction, random effect of participant, and dependent variable of reported sensation (vibration versus pressure). Overall, participants were significantly more likely to experience vibration when stimulation frequency was higher (GLMM fixed effect 𝛽 𝑙𝑜𝑔(𝑓) = 1.31, 𝑆𝐸 𝛽 𝑙𝑜𝑔(𝑓) = 0.33, 𝑊𝑎𝑙𝑑 𝑍 = 3.96, 𝑝 < 0.001). On average, increasing the stimulation frequency by 70% doubled the odds of experiencing vibration (versus pressure), and the model predicts that the "crossover" point at which participants first begin to report vibration more often than pressure is ~6𝐻𝑧. In contrast to the effect of stimulation frequency, stimulation current did not significantly influence the type of sensation reported by participants (GLMM fixed effect 𝛽 𝑙𝑜𝑔(𝐼)×𝑙𝑜𝑔(𝑓) and 𝛽 𝑙𝑜𝑔(𝐼) terms; both 𝑝 > 0.2).
The Intensity rating for changes in stimulation current was assessed. Whereas changes in stimulation frequency produced qualitatively different sensations (pressure versus vibration), changes in stimulation current seem to produce quantitatively different sensations (more versus less intense). To quantify this observation, data were fit using a linear mixed effect regression (LMM) with fixed effects of log stimulation frequency, log stimulation current, and their interaction, random effect of participant, and dependent variable of reported intensity (1-5 scale).
Overall, the sensation produced by stronger currents was rated as significantly more intense (LMM fixed effect 𝛽 𝑙𝑜𝑔(𝐼) = 0.36, 𝑆𝐸 𝛽 𝑙𝑜𝑔(𝐼) = 0.05, 𝑡(176.01) = 3.18, 𝑝 = 0.002); on average, doubling the stimulation current increased the intensity rating by approximately 0.25 (on the 1-5 scale).
Next, we assessed the amount of discomfort during stimulations. Importantly, all sensations were reported to be between "Very comfortable" and "Neutral" except for a single trial for a single participant (who rated 10 Hz stimulation at detection threshold as "Uncomfortable").
In addition, we assessed the localization of the perceived stimulation. Almost all sensations (93%) were experienced as localized rather than radiating (5%) or referred (1%). To quantify the relationship between stimulation frequency, stimulation current, and reported location, data were fit using a logistic mixed effect regression (GLMM) with fixed effects of log stimulation frequency, log stimulation current, and their interaction, random effect of participant, and dependent variable of reported location (localized/not localized, dummy coded as 1/0).The probability of experiencing a sensation as localized did not significantly depend on frequency or current (GLMM fixed effect 𝛽 𝑙𝑜𝑔(𝐼)×𝑙𝑜𝑔(𝑓) and 𝛽 𝑙𝑜𝑔(𝐼) terms; both 𝑝𝑠 > 0.05). Overall, participants experienced the sensations as localized far more often than would be predicted by chance (GLMM estimated marginal mean, 𝑃(𝑙𝑜𝑐𝑎𝑙𝑖𝑧𝑒𝑑) > 0.99, 𝑆𝐸 = 0.009, 𝑊𝑎𝑙𝑑 𝑍 = 2.78, 𝑝 = 0.005).
To quantify participants' ability to discriminate between two different stimulation currents, data from the 2AFC discrimination task were fit using a logistic mixed effect regression (GLMM) with fixed effect of difference in (log-scaled) stimulation current, random effect of participant, and dependent variable of trial outcome (correct/incorrect). Overall, participants were significantly more accurate when the two stimulation currents were more different (GLMM fixed effect 𝛽 Δlog(𝐼) = 5.92, 𝑆𝐸 Δlog(𝐼) = 1.00, 𝑊𝑎𝑙𝑑 𝑍 = 5.94, 𝑝 < 0.001); on average, each 13% increase in the difference between stimulation currents doubled the odds of detecting the difference.
Furthermore, the smallest difference for which 𝑃(𝑐𝑜𝑟𝑟𝑒𝑐𝑡) > 0.75 (specifically, the psychophysical threshold for discrimination) was 1.33 times the detection threshold (as estimated by GLMM estimated marginal mean, 𝑃(𝑐𝑜𝑟𝑟𝑒𝑐𝑡|Δlog(𝐼) = 1.33) = 0.75).
For the comparison study, we have used Ag/AgCl electrodes (EL507, BIOPAC Systems, Inc.) with electrode gel (Spectra 360, BIOPAC Systems, Inc.).