HaptiCoil: Soft Programmable Buttons with Hydraulically Coupled Haptic Feedback and Sensing

Providing clear haptic feedback during both pressing and releasing actions enhances user experience by offering unambiguous system confirmation. Haptic devices can generally be categorized into passive and active devices. Passive devices offer haptic feedback by dissipating mechanical energy applied by the user, through spring mechanisms. For over a century this has been the dominant way to provide button feedback, and a range of microswitchs, capacitive membrane switches, and every manner of scissor mechanism has been implemented in keyboards, mice, and gaming controllers. Some of these mechanisms have been examined in an HCI context, such as [28, 73]. However, once constructed, passive haptic button response is unvarying.

Active haptic devices, on the other hand, provide controllable haptic feedback through external energy sources. Piezoelectric actuators, such as the commercially available TDK Powerhap and Piezohap actuators, are promising for haptic applications due to their small size, fast response, and low energy consumption [4, 32, 57]. Despite challenges such as small displacement and the need for high voltage (50-200V), companies like Boreas Technologies have introduced efficient piezo drivers that could solve the issue, though at a cost. Other compact haptic actuation methods have also been explored, the most common including shape memory alloys [13, 22, 35], electrotactile actuators [29, 61], dielectric elastomer actuators (DEAs) [17, 24, 69, 70]. The typical limitation of these technologies is their small output mobilities (ability to move large distances quickly) as well as their commercial readiness and acceptance. Even now, however, SMA and piezo based buttons are making it into certain mobile devices due to their extreme compactness.

Thus, electromagnetic actuation, commercialized for over a century, remains one of the most common methods to provide haptic feedback. ERMs and LRAs offer strong vibrotactile feedback with high efficiency, with the trade-off of poor output control (ERMs) or narrow frequency range (LRAs). Various miniaturized haptic devices have been developed using commercially available ERMs [49] and LRAs [8, 20, 42, 58]. However, their limited frequency range restricts their use in more advanced haptic applications requiring "high-definition" output, and many times their form factor does not match the intended use case. To address this, researchers have also developed their own electromagnetic actuator integrations [34, 40, 45, 62]. Vibrotactile coil based products, such as Haptuator from Tactile Labs [65], and C-2 from Engineering Acoustics Inc [37], have also become the gold standard for vibrotactile research due to their broad operating frequency range (10-1000 Hz). However, their high cost ($80-$200) remains a barrier to widespread experimentation. So-called "tactile-exciters", a class of coil based actuators originally designed for bass reproduction applications, have also been used in haptics research, though typically in configuration similar to LRAs, therefore we treat them as a similar class. These devices are typically bulky and offer limited customizability, see Figure 2 for a full actuator comparison chart (categories adapted from [7]). Finally, wearable haptic devices utilizing miniature DC motors also show great potential for providing multi-DOF tactile feedback, yet are typically reserved only for static or low frequency force rendering, see [39] for a review.

Matrix showing pros and cons of haptic actuators

A variety of haptic devices, including haptic buttons, shape changing devices, and wearablables, have been recent explored using fluidic (e.g. pneumatic or hydraulic) actuation.

Pneumatic systems offer strong advantages, including being lightweight and delivering high power performance. However, the need for bulky air compressors, valves, solenoids, and air supply lines to control them complicates miniaturization efforts [56]. One of the most common approaches in pneumatic-based haptic devices involves inflating and deflating soft silicone membranes using pneumatic control lines. It is so common that HCI toolkits have also been developed for easing implementation [52]. This method has been applied in dynamic haptic buttons [15], shape changing devices [66], braille devices [16, 47, 63], tactile feedback devices [6], and wearable haptic devices [31, 33, 71, 72]. Commercial approaches remain extremely high cost, and recent works have aimed developing compact pneumatic-based haptic devices by replacing external air compressors with voice-coil actuation [50, 53], electrostatic actuation [55] and even thermal actuation [60]. However, their low output force or slow actuation speed remain as a challenge.

Hydraulic actuation holds great potential for compact haptic devices due to its fast, stable, and reliable power transmission capabilities [2]. Moreover, hydraulic actuators provide good mechanical impedance matching with human skin and underlying tissue (being primarily composed of water), making them more attractive methods for haptics. The simplest hydraulic approach involves using commercially available hydraulic pumps and valves. Wearable rings [14] and pens [64] have been developed using these methods. Very recently, researchers have focused on developing compact, lightweight advanced hydraulic pump techniques, such as electrostatic zipping actuation [5, 25, 43], hydraulic coupled dielectric elastomer actuators [68], and electroosmotic pumps [44, 51, 54]. Despite these significant advancements, reliability and stability in the long term remain understudied. A notable alternative approach is described in [36], which uses human input for pumping, though this work was focused more on visual interactions rather than controllable tactile feedback. A comprehensive comparison between fluidic haptic approaches and the HaptiCoil is presented in Figure 2.

The haptic button is an I/O interface that provides both haptic feedback and touch sensing. A simple method to achieve sensing capability is by combining haptic actuators with external sensors, such as force sensor [5, 23], accelerometers, [11] or microphones [59]. Recently, several attempts have been presented to develop self-sensing haptic actuators that integrate both tactile sensing and haptic feedback.

Piezoelectric actuators offer self-sensing capabilities by detecting the charge generated from piezoelectric effect, an effect demonstrated in [67]. DEAs also posses sensing capabilities by monitoring changes in capacitance [46]. A few researchers have presented DEA-based haptic devices with self-sensing functionality [12].

Electromagnetic actuators again are particularly promising for self-sensing, due to a strong coupling between electrical and mechanical domains. Recent research has shown that self-sensing LRAs can utilize back electromotive force (back-EMF) to detect change in velocity of the internal mass under applied pressure [9]. While these device can measure external pressure, reliable measurement of back-EMF requires disconnecting the actuator while sensing, which complicates the control circuit. Compared to back-EMF sensing, self-sensing haptic devices that actively measure low frequency current through the voice-coil can effectively detect applied pressure [8, 27, 38]. Back-EMF and low-frequency current sensing are inherently tied to the actuation waveform, requiring significant voltages (>1 V) to sense mechanical changes like damping or stiffness. This coupling limits waveform options and precludes sensing without significant actuation. For HaptiCoils, we developed a low-voltage (20 mV) sensing method at high frequency (>1 MHz), frequency decoupling it from tactile actuation (<500 Hz). Unlike back-EMF and current-sensing, which struggle with combining static (DC) and highly dynamic (>100 Hz) sensing, our high-frequency self-inductance measurement effectively detects static and rapidly changing coil displacement (and by proxy, applied force), representing what we believe is a novel and significant contribution to the field. Our method does not yet allow for truly simultaneous sensing and actuation, but we believe there may be a path to allow this.

HaptiCoils are constructed from waterproof miniaturized speakers which are coupled to a soft silicone membrane via a fluid, as shown in Figure 3 A. This design combines the speaker's high bandwidth with the fast and reliable power transmission of the hydraulic system due to the incompressible nature of the fluid.

Image showing the exploded view of HaptiCoils, and also how they are filled and appear once completed.

The fabrication process of our Hapticoils is presented in Figure 4. The fluid chamber is prepared by 3D-printer (Formlabs Form 4) with clear V4 resin (Formlabs). The transparent resin allows for easy inspection of trapped air bubbles, although the chamber can also be fabricated using other 3D printing materials, such as Acrylonitrile Butadiene Styrene (ABS) or Polylactic Acid (PLA). The fluid chamber is designed to be slightly larger than the speaker dimensions (chamber size= 18 × 13 × 3 mm in this case) and includes a small hole (1 × 2 mm) at one edge to facilitate fluid injection. Then, the fluid chamber is bonded to the speaker with cyanoacrylate adhesive (Loctite 401) to prevent fluid leakage. The soft touch membrane, which serves as a pressure-coupling material between the speaker and the user's skin, is a pre-cured material (Rogers HT-6000, Shore 10A, 0.25 mm thickness) purchased from McMaster-Carr. Adhesion of the silicone membrane to the fluid chamber can be difficult, but we found two reliable methods. First, a plasma surface treatment wand (Electro-Technic Products BD-20A) can applied to the soft membrane to activate its surface. The treatment wand is passed over the membrane for several seconds, then it is adhered using laser-cut pressure-sensitive adhesive (PSA) sheets (3M 467MP). Second, a cyanoacrylate adhesive (Loctite 401) can be used directly between the fluid chamber and membrane. We found the former processes more reliable and clean, and the latter process is cheaper but messier. After bonding, water is injected into the chamber, and air is evacuated using a syringe and dispensing needle through a small filling hole, which is designed in the side of the chamber wall, see Figure 4 B. We use tap water as the fluid due to is low viscosity, low toxicity, and easy of use. Finally, the filling hole is sealed with room temperature vulcanizing (RTV) silicone adhesive (Permatex 80050).

Image of a linear lumped impedance model of speakers, and an image showing a range of sizes and shapes of speakers we tested.

To aide in selecting a speaker we pre-screened various shapes and sizes of speakers (Figure 5 B) by measuring the Thiele/Small (T/S) parameters using a two-channel FFT analysis, similar to that used in [53]. Additionally, speakers were required to have a high IP water resistance rating. Figure 5 A presents a lumped parameter model we fit to each speaker. We selected the miniature speaker (CMS-151135-18S-X8) that has the highest BL parameter (also known as force factor) of those tested (1.22 N/A). High BL implies large output forces for a given current, and we found this to be the single most useful parameter in selection. We also sought to find speakers with low coil resistance, allowing us to apply large currents for relatively small voltages. Using this approach additional speakers on the market can be quickly pre-screened for their appropriateness in HaptiCoils.

Diagram showing direct and indirect actuation schemes.

Hydraulic systems offer a distinct advantage in efficiently transmitting power from one point to another. Leveraging this advantage, a HaptiCoil can be designed using both direct and indirect actuation, as illustrated in Figure 6 A, B. In the direct case, pressure is straightforwardly focused on the touch surface. In the indirect case, the touch surface is physically separated from the actuator via a hydraulic channel. This separation affords HaptiCoil designers more flexibility to create complex structures or a dense array within a limited space. An example of this in shown later in Figure 21.

We also found that HaptiCoils are adept at providing vivid output shapes than can quickly pop up or down from the surface of the device. We believe this can lead to interesting haptic and visual output designs for shape display purposes and emphasize this point by showcasing various shaped outputs. These were achieved by changing the shape and size of HaptiCoil surface hole. Figure 6 C shows some representative shapes in the popped out configuration. During this test, we applied a square wave input a 200 ms period. Images were captured with a high-speed camera (CHRONOS 2.1-HD). Pop down shapes can be seen in Figure 22. Due to the soft nature of the membrane, the edges of the output hole can also be felt by users as a designable passive haptic effect.

Graph of sensor output voltage changing with respect to speaker cone displacement, and an image showing the sensing principle in action

Sensing of applied pressure in our HaptiCoil prototypes is done through a high frequency (> 1 MHz) self-inductance measurement. Through our testing, we observed that the self-inductance of the speaker coil systematically changes as it moves inwards (positive displacement) or outwards (negative displacement) due to an externally applied force. Specifically, the self-inductance decreases when the cone moves inwards and increases when it moves outwards. We measured this phenomenon using an LCR meter operating at 200 kHz. The exact mechanism driving these inductance changes remains unclear without detailed knowledge of the coil's internal geometry and the structure of the magnetic field. We hypothesize that these inductance variations result from changes in the magnetic flux through the coil as its position shifts. If the coil moves away from the flux lines, self-inductance would decrease, which aligns with observations of similar microspeaker constructions. Furthermore, since cone displacement is hydraulically linked (Figure 7) to the displacement of the output surface, our HaptiCoil prototypes are very sensitive to surface displacement and pressing force as well.

An electrical schematic of HaptiCoil electronics, and a photo of PCBAs and HaptiCoil button

To measure the self-inductance of the speaker while maintaining the haptic operation of the speaker, we needed to build a custom control circuit (shown in Figure 8 A) which includes both actuating and sensing electronics in a single signal path. This circuit includes a small differential Class-AB audio amplifier (TI TPA6211A1), a shunt inductor, and a dual channel pre-amplifier (TI OPA2354) consisting of a high gain differential stage followed by a peak detector. A 3 MHz sense signal is injected by our micro-controller, and is added to a low frequency (<1000 Hz) haptic signal using an additive input configuration. It should be noted that the user can not feel the vibration generated by the sense signal, due to its small amplitude (20 mV) and high frequency (3 MHz). In-practice, we time multiplex these signals, meaning we are only sensing or actuating at a given time, not in combination.

Considering only the sensing path, the sensing oscillation signal is injected through the same audio amplifier that plays the haptic signal. The amplifier load is modified to include a shunt inductor in series with the speaker, as seen in Figure 8 C. Using an inductor instead of a resistor means that the shunt inductor adds minimal additional impedance at haptic frequencies. At our sensing frequency (3 MHz), however, the shunt inductance and speaker's variable self-inductance form an inductive divider, and the amplitude of the signal at Vsense is thus highly sensitive to changes in speaker self-inductance. The rest of the circuit is designed to convert this high frequency sinusoid into a rectified DC signal. The voltage is amplified using a differential op-amp with G ≈ 100, and a peak detector creates an easy to read signal for the micro-controller ADC. The sensor output has a inverse relationship with the self-inductance of the HatptiCoil due to the inductor divider relationship.

Cost is a primary consideration in our design. The main cost savings comes from the low cost of the speaker ($1.3). Remaining materials are inexpensive: silicone membrane (23¢/g), 3d printed resin (20¢/g), adhesive sheet (0.001¢/mm2), and silicone adhesive (5¢/g), which makes total cost less than $1.5. The price of the electronic sensing and actuating components per channel is less than $3.5, while our MCU board (Teensy 4.0) is a large one time cost ($23.8). We anticipate that the cost of driving electronics could come down significantly with scale, as they are all low voltage (<5 V), and are similar to the electronics in mass-produced smart style earbuds.

One of the most important aspects of a haptic button is the output force it can provide to the skin, specifically the skin of the fingertip. Previous studies have shown that the typical pressing force during tapping a touchscreen, typing on a physical keyboard, and clicking a mouse button is around 0.5 N [1, 18, 21]. Thus, our goal for HaptiCoil was to deliver sufficient output force under this amount of preload.

Image showing blocked force measurement setup

Data plots of blocked force recordings

We measured blocked force using a piezoresistive strain gauge force sensor (ATI Nano17) and a fingertip-like proxy. The blocked force refers to the force generated by the HaptiCoil when its motion is completely restricted by the force sensor. The initial contact force between the HaptiCoil and the force sensor was precisely controlled using a motorized linear stage (Optics Focus MOX-02-100). We attached fingertip-like structure covered with soft foam resembling human skin to serve as a coupling between the force sensor and the HaptiCoil. A data acquisition device (NI USB-6211) generated haptic test signals various frequencies and simultaneously measured the force sensor data and the current flowing through the speaker. We tested three different HaptiCoils with output hole sizes ranging from 4 mm to 6 mm. The test setup can be seen in Figure 9.

Data plots of blocked force recordings

Graphs showing force recordings over time, including step response and sinusoidal inputs

Figure 10 A shows the profile of the blocked force of the HaptiCoil at various operating frequencies (preload = 0.5 N). Results validated that our HaptiCoils can operate at all frequencies tests, from 1-400 Hz, as can be seen in Figure 10 B. The HaptiCoil with a larger hole size produces a larger output force as the output force is proportional (F = PA) to the area under the same applied pressure in a hydraulic system. We also validated that our HaptiCoil (hole size: 6 mm) can produce a controllable output force by controlling the input power, shown in Figure 10 C. To verify whether the output force of HaptiCoil is sufficient for a haptic button applications, we estimated the resulting skin deformation using a dynamic model that couples HaptiCoil with the fingerpad skin, as can be seen in Figure 11 A. The results presented in Figure 11 B demonstrated that skin deformation is at least 4 times larger than the reported detection threshold at low frequency (1 Hz), and up to 200 times larger for high-frequency vibration stimuli (> 100 Hz) [3]. Results shown in Figure 11 C further validated that our HaptiCoils can deliver perceptible force even under a high preload of 10 N. This is achieved because the hole structure limits the pressure transmitted to the internal fluid, effectively preventing the collapse of the speaker membrane. Figure 12 A shows that we can produce a stable output force during 1,800 cycles (10 s operation) at 180 Hz. The step response of a HaptiCoil is shown in Figure 12 B, with calculated rise and fall times of 2.1 ms and 2.6 ms. This fast speeds validate the effectiveness of using an in-compressible fluid for transmitting force.

Photo of a displacement sensor measuring HaptiCoil free displacement.

Graphs of free displacement at various frequencies and with different HaptiCoil hole sizes.

Images and graphs showing displacement on two differnt locations of a HaptiCoil, the edges and the center

The surface deformation is also a crucial aspect of a haptic button, as it directly impacts haptic perception. Unlike the blocked force test, we measured the displacement of HaptiCoil without any external preload. To effectively stimulate the fingerpad, the required displacement of the haptic device varies depending on size and operating frequency of the device. In practical use, the output displacement of the HaptiCoil is less than free displacement due to the load from the skin (as described by our dynamic fingertip model), however these recordings still offer a good benchmark of performance. We recorded the output free displacement of HaptiCoils at various operating frequencies using a photoelectric sensor (Baumer CH-8501) and a data acquisition device (NI USB-6211). The focal point of the laser was controlled using a linear stage. Figure 13 shows the displacement measurement test setup. A sinusoidal signal was applied for 5 seconds, and the displacement and current were recorded simultaneously.

The displacement profiles of three HaptiCoils at various operating frequencies (1-400 Hz) are presented in Figure 14 A. During the experiment, the average input power was maintained at 2.7 W (current= 1 A). The results indicate that the HaptiCoil with a 5 mm hole size produced the largest displacement in low frequency range (1-60 Hz), with an estimated -3 dB bandwidth of approximately 70 Hz. Figure 14 B shows the peak displacement curve of a HaptiCoil (hole size = 5 mm) under different input power at a 10 Hz operation. Calculated rise and fall times to a step input are 14.7 ms and 19.4 ms, respectively, indicating rapid inflation and deflation speeds (see 14 C). Additionally, we compared the displacement at the center and edge of the HaptiCoil, as shown in Figure 15 A, to confirm that our actuator can produce pure localized displacement. The results in 15 B confirm that our actuator can produce a localized vibration at the surface with minimal displacement outside the touch surface, a result that is also felt qualitatively.

A graph of sensor data versus applied force and realtime sensing data.

A graph of sensor data versus applied force and realtime sensing data.

HaptiCoils can measure input force by monitoring the inductance of the coil, as introduced in Section 4.4. In Section 4.5, we presented the circuit to measure the inductance of the coil in real-time. Figure 16 A presents the sensed output profile of a complete HaptiCoil as a function of applied force. For this test, we used a motorized linear stage and a force sensor, identical to those used in the blocked force measurement test. The results demonstrated remarkably linear relationship between the sensor output and applied force within the range of 0-1 N. Figure 16 B compares the sensor output of the HaptiCoil to the reference force captured by the force sensor under touch input conditions with a real finger. Results demonstrated that the HaptiCoil could rapidly detect pressing forces with a 5 ms delay and releasing force with a 100 ms delay. We further investigated sensing performance of the HaptiCoil under an energized button-click scenario. The HaptiCoil was designed to provide click feedback during both the pressing and releasing phases of button interaction, as shown in Figure 17. A second-order digital biquadratic low-pass filter with a cut-off frequency of 1000 Hz was applied to the sensor output to reduce high-frequency vibration transients. Results showed that our self-inductance measurement is significantly affected by the applied impulse haptic signal. We hypothesize three main reasons for this, including: rapidly changing coil position (our main sensing effect), changes in flux due to large applied current (which we have measured by blocking coil displacement), and the limited common-mode rejection (CMRR) of our sensing pre-amplifier. Low frequency CMRR is also limited due to the small remaining resistance of our sense inductor (0.23 Ω). We believe these effects could be reduced or compensated for in future work, however to address this interference in this implementation sensor output was temporarily disregarded immediately after applying the haptic signal. This allowed for stabilization over a 100 ms period. While further investigation is needed to explore potential compensation methods, the findings confirmed that HaptiCoils can accurately monitor real-time, consecutive button presses at a high speed of 200 ms intervals. Furthermore, the HaptiCoil exhibited rapid response to sensor output, achieving a 0.8 ms delay and generating a 1.25 ms long haptic click.

Due to their lightweight and compact nature, HaptiCoils are well suited for wearable applications. We built a small, indirect type HaptiCoil with the speaker mounted to the fingernail and the touch surface wrapped around the tip of the finger (Figures 20 A, B). Due to the sensitivity of the HaptiCoil's input, it can sense being put on, as this causes a slight deformation in the touch surface. Once fitted, the device can not only provide localized haptic feedback directly to the fingerpad (Figure 20 D), it can also detect force between the bottom of the case and the outside world, as shown in the scenarios seen in Figures 20 E, F, and G.

First, imagine an VR number input scenario, Figure 20 E, where a number pad is revealed when a user makes a typical pinch gesture (as detected by the head mounted device's on-board cameras). Now imagine the user can move their hand (while still holding the pinch gesture) and select a specific number from the group, confirming their selection by "clicking" on it via a force pinch. This confirmation is detected by the HaptiCoil and click feedback profiles clear communication to the user of their selection.

Next, a number pad can also be co-located with real physical surfaces, such as a workstation desktop, Figure 20 F. As the user's finger nears the desktop, the on-board cameras project their position to highlight an intended number. Once the user contacts the surface, they feel kinesthetic feedback from real grounded reaction force. If they continue pushing against this force, the HaptiCoil can detect an input force and trigger a click confirmation.

Finally, imagine a self-haptics scenario, where a number pad appears on the user's opposite hand once they turn their palm towards the on-board cameras. Similar to the grounded desktop, a user can push their finger against a specific location of their hand, selecting a button and receiving force appropriate click feedback to confirm their selection. In each of these 3 scenarios, the HaptiCoil serves as a wearable button, taking on different functions depending on the context of user interaction. In this sense, it a type of spatial computing equivalent of a traditional mouse-click. Of course, HaptiCoils can also be triggered due to totally virtual buttons, such as in Figure 1 C.

We also envisioned and built a 3D printed digital pen, complete with a conductive rubber pen tip, and two integrated counter firing HaptiCoils. The speakers were mounted near the rear of the device, and the output was routed to two small openings (4 × 4 mm) (seen Figure 21 A-C). This rear actuator configuration is similar to the latest Apple Pencil (which includes vibrotactile feedback); however, we use hydraulic channels to focus localized haptics into the compact space between the fingers near the tip of the pen, as it is difficult to place speakers in that area directly. Using hydraulic channels, the digital pen can achieve a compact structure similar to that of a commercial digital stylus, as shown in Figure 21 C.

We designed an example drawing application on a Surface Go tablet, as shown in Figure 21 D, to demonstrate haptic input and output. Single click input Figure 21 E is used to quickly switch between drawing and erasing modes, while a force pinch and drag "double-click" input is used to change stroke width (Figure 21 F). We also envision directional cues rendered to the user, by firing phase offset haptics to the index finger and thumb (Figure 21 G), a technique known to create apparent tactile motion. This motion can help indicate, for example, the rightward or leftward highlighting of text while reading an article. Finally, broadband tactile textures can be rendered via our HaptiCoils, depending on the virtual texture the pen is sliding across on screen, Figure 21 H, adding a sense of realism to the digital inking experience.

Images of a remote control arrayed example.

For our final example, we built a low-profile programmable click remote control (total thickness of 7 mm, outside dimension 45 × 45 mm), as shown in Figure 22. This controller housed an array of 5 separate HaptiCoils, each sensed and driven independently. We designed the HaptiCoil output hole shapes to match an up, down, left, right, and center configuration, allowing users to feel the locations of the buttons without looking (Figure 22 A). We found that implementing multiple HaptiCoils in close proximity created minimal interference, however we could detect some subtle cross-talk in self-inductance baselines for particularly strong haptic effects. This effect, however, could be easily compensated, or even ignored, as buttons were only clicked one at a time. The use case for this example is a media player, where a user can pause or play the song by clicking on the center button, and skip forward or backwards using the right and left buttons. Clicking up and down allows them to change between songs. We implemented keyboard HID commands to emulate a keyboard media player, and programmed custom click sensations in our micro-controller.