Overview
NaptX puts a vibrotactile actuator on the back of the fingernail instead of on the fingerpad. The pad stays free, so you can touch a screen, a key, or a real object while receiving haptic cues about it.
It exists as two devices built for two different jobs: NaptX, a self-contained wireless unit that lives on a single nail and is optimised for size, cost and wearability; and Napt10X, a wrist-worn research platform that drives ten actuators, reads ten accelerometers, and exists to run experiments quickly. Both came out of my PhD at Imperial College London.
Why the nail
Haptic gloves and fingertip devices deliver rich sensations by covering the one surface the brain relies on most. Bracelets and body suits leave the hands free but can't create the illusion of touching anything with the hand.
The fingernail sits between those two failures. It is rigid, coupled almost directly to bone, and surrounded by a dense field of tactile receptors - so it transmits vibration efficiently and a small actuator there is clearly felt. The catch is that stimulation on the nail can feel like a foreign object rubbing against it, which is exactly what continuous buzzing from an ERM motor does.
Short impulses from an LRA behave differently. Their origin is ambiguous - they can be read as coming from the fingertip itself. The whole project is a bet on that ambiguity: if the brain can't reliably tell nail from fingertip, feedback on the nail can be made to feel like touch.
NaptX - the single-nail unit
- 9 × 6 × 15 mm, 2.45 g - light enough that inertia doesn't tug on the nail during fast hand movement
- Origami-folded flex PCB wrapping around a coin LRA, bonded to a 3D-printed support that sits on the nail - the video above shows the KiCad flex viewer written to preview the fold in 3D before ordering
- ESP32-C3 for control and Bluetooth, with the antenna routed directly on the flex - reliable wireless at this size was one of the harder problems
- STSPIN250 current driver for precise amplitude and frequency control
- 3-axis accelerometer for finger motion, and an RGB LED plus push button so it can be used without another interface
- 150 mAh LiPo - several minutes of continuous vibration, aimed at short-session use
- Roughly £20 in parts
Napt10X - the research platform
Where NaptX is one channel done small, Napt10X is ten channels done flexibly.
- Teensy 4.0 for real-time waveform generation and 480 Mbit/s USB, with an ESP32-C3 for WiFi and Bluetooth
- 10 × Vybronics VG0640001D LRAs - 6 mm × 4 mm, 1 g, up to 1.3 G at their 230 Hz resonance, Q-factor around 5
- 5 × STSPIN250 drivers at 1.3 A, PWM at 50 kHz to stay out of both the audible and the tactile range, each with a low-pass output filter
- Two 8-bit shift registers collapse the 20 direction inputs down to 12 microcontroller pins
- 10 × LIS3DH accelerometers, multiplexed and read at 1 kHz, so the actual delivered vibration is measured rather than assumed
- SGTL5000 audio codec (16-bit, 44.1 kHz) and an EYESPI header for a small display - hooks for multimodal experiments that haven't been used yet
- Around £55 in parts, or about £90 with the audio and display options
Driving an LRA properly
An LRA is a magnet on a spring inside a coil. It is cheap, small, and a poor actuator for anything except a sine wave at its resonant frequency - the Q-factor above means that at 115 Hz or 345 Hz you get roughly half the amplitude for the same power. Getting controllable feedback out of it is the engineering problem.
Drive chain. PWM into an H-bridge driver, then a second-order LC filter (33 µH, 1 µF, corner around 5.7 kHz) to turn the switching waveform into something the actuator's tiny moving mass responds to linearly.
Model. Electrically the LRA is a 25 Ω, 1 mH coil in series with a back-EMF proportional to the velocity of the moving mass. Mechanically it is a mass-spring-damper driven by the coil force. Coupling the two gives a transfer function from filtered voltage to displacement that is a second-order resonance in series with an electrical first-order lag - which is the narrow passband you have to control around. The full derivation is in the repository README.
Does the brain know where it came from?
The first user study tested the central assumption. Fifteen LRAs were taped to one hand - nail, fingertip, and medial phalanx on each of five fingers - and 18 participants localised and rated the intensity of single pulses over 90 trials.
| Stimulated | Felt at fingertip | Felt at nail | Felt at knuckle |
|---|---|---|---|
| Fingertip | 43% | 49% | 5% |
| Nail | 17% | 65% | 7% |
| Knuckle | 3% | 11% | 75% |
People could not reliably tell nail from fingertip. Stimulating the fingertip was more often reported as the nail than as the fingertip. The knuckle, further away, was localised fine. The ambiguity is real, and it is the thing to build on.
This study and the next were run with Qi Li, an MSc student I co-supervised with Yanpei Huang and Etienne Burdet.
Feeling a remote surface through your nail
The second study asked whether that ambiguity is useful. A Kinova Gen3 arm carried a sensorised probe over silicone-covered boards with hidden relief underneath - the surgical analogy is a polyp under tissue. The friction-induced vibration picked up by the probe was replayed on the participant's nail, and the same finger's orientation drove the arm.
Three tasks of rising difficulty: rank the density of hidden bumps in four corners, locate four hidden squares on a 3 × 3 grid, and identify hidden shapes. Each was run twice - once with the participant teleoperating the arm, once with the arm scanning automatically - to test whether controlling with the finger you're also sensing through gets in the way.
It didn't. Accuracy in teleoperation was slightly higher in all three tasks (not significant), and the hardest task was completed faster under teleoperation. Most participants reported that actively controlling the exploration helped their perception. Using one finger to both act and feel, with nothing covering the fingerpad, works.
Demonstrations
Bump illusion. An Android app shows a grid of dots; sliding a finger over the screen fires a pulse on the nail at each dot. Nearly everyone reports feeling bumps at the fingertip, not the nail. The more telling observation: if someone else touches the screen while the wearer holds their finger in the air, the sensation is artificial and unpleasant - but if the wearer moves their raised finger in sync, some feel a bump anyway. Context and agency shape where a sensation lands.
Remote texture. Two NaptX units, one on each index finger. One records the natural friction vibration of sliding over a surface, band-pass filters it, and streams it to the other, which replays it on the opposite nail. Participants described it as realistic, particularly when the passive finger was resting on a surface.
Sensing force through the nail
The missing input is fingertip force - vibration propagation through the finger depends heavily on how hard it is pressing, so matching feedback to a touch needs to know that. Most force sensors go where the touch is, which defeats the purpose.
The approach being explored uses the nail itself as a dielectric: a thin copper plate on the device forms a capacitor with the fingertip through the nail, and pressing deforms the nail enough to change the capacitance measurably. Early prototypes give a clean real-time signal as finger pressure oscillates. Fringe effects and robustness across users are the open questions.
What's still open
Roughly half of nail stimuli are still perceived at the nail. On its own that ambiguity isn't enough - real use needs the fingertip reading to be consistent. The demonstrations point at the answer: giving the brain a convincing context, such as a finger visibly touching a screen, strengthens the illusion considerably. Adaptive feedback and multimodal cues are the next things to try.
Student work on this platform
Two MSc projects at Imperial are building on NaptX this year - one using it to cue finger sequences for piano training, the other characterising a piezo-driven variant for rendering texture.
Open hardware
Everything is in the repository: KiCad schematics and layouts for both boards, the bill of materials with supplier part numbers, a C++ library, and a command-line demonstration app.
Status
Active development as a haptics module of the CleverHand platform. Current work is on a new actuator design.