InferTouch

A sealed, air-filled robotic fingertip designed to help a human-like hand “feel” material properties: contact pressure, vibration/acoustics, and temperature—via a single I²C (Inter-Integrated Circuit) bus on a flex tail.

sealed air cavity for ΔP sensing

embedded skin thermistors

pressure + mic + ambient temp on PCB

single I²C bus via flex

Datasheet Overview

InferTouch fingertip sensor concept render

What it is

InferTouch is a sealed fingertip module with a compliant silicone skin. Temperature sensors are embedded in the skin, while an internal PCB carries a pressure sensor, microphone, and ambient air temperature sensor. The internal cavity is air-filled so that contact events generate measurable pressure deltas. All devices share a single I²C bus that exits as a flex PCB.

Datasheet (PDF)

Interface, timing, mechanical envelope, environmental limits.

Integration notes

I²C addressing, sampling, filtering, and calibration workflow.

Use-cases

Slip detection, texture classification, thermal response, contact onset.

Sensing stack

Skin temperature: multiple embedded sensors in silicone for surface thermal gradient.
Internal pressure: sealed air cavity converts contact deformation → ΔP at pressure sensor.
Vibration/acoustics: microphone captures high-frequency content (texture / slip / impacts).
Ambient air temperature: internal reference to compensate thermal drift and gas law effects.

Electrical interface

Bus: single I²C shared by all sensors (pull-ups on host or on fingertip, depending on variant).
Flex tail: I²C + power exits from fingertip base as flex PCB for routing inside a robotic finger.
Addressing: plan address map (some devices fixed; others configurable via strap or mux if needed).
Sampling: pressure + mic are typically higher-rate than temperature; decimate/aggregate in firmware.

Typical use-cases

Contact onset & force proxy: detect touch and estimate applied force from cavity pressure response.
Slip detection: microphone bandpower + pressure micro-oscillations indicate incipient slip.
Texture classification: spectral features from vibration signal during scanning motion.
Thermal properties: heating/cooling transient at skin sensors suggests conductivity / heat capacity.

Data products you can expose

Pressure: raw ΔP, filtered ΔP, contact event timestamps, step response parameters.
Vibration: RMS / bandpower (e.g. 50–300 Hz, 300–1 kHz, 1–5 kHz), transient markers.
Temperature: skin map, gradient, ambient reference, drift-compensated features.
Fusion: “slip score”, “texture signature”, “material class” (if you run a model).