SoftShell
Technology

A complete tactile stack — sensing, signal, and pipeline.

SoftShell combines conformable tactile sensing, low-noise signal capture, and integrated fast processing into a single deployable infrastructure layer. The result is a tactile system that scales from human demonstration to real robotic deployment without changing hardware.

Form factor
Conformable film
Capture
Low-noise
Pipeline
Integrated
Deployment
Glove + Skin
Overview

Tactile infrastructure, not a single component

Most flexible sensors solve one slice of the problem — coverage, fidelity, processing or deployment. SoftShell is engineered as a complete stack so the same sensing layer remains useful all the way from demonstration to deployment.

SS-MESH / CONFORMABLE FILM

v0.2 SAMPLE

Architecture

From contact to actionable signal

The SoftShell stack is engineered around four functional layers — each one solving a specific failure mode of conventional flexible tactile sensing.

L-01

Sensing layer

A conformable, low-profile film designed to wrap real geometry — fingers, palms, forearms and end-effectors — without compromising fit or feel.

L-02

Signal quality

A low-noise capture architecture targeting the fidelity required for precision manipulation, instead of trading signal quality for conformability.

L-03

On-sensor processing

Integrated processing at the sensor level so signals are conditioned and encoded close to the source — reducing host-side bottlenecks.

L-04

Fast data pipeline

High-throughput transmission engineered for the rates real manipulation requires, so the tactile layer never becomes the bottleneck of the platform.

System architecture diagram · v2

glove ↔ skin · single sensing format

SS-ARCH / TACTILE STACK · v201Conformable Sensingflexible film · full-surface coverage02Low-Noise Capturesignal conditioning · fidelity-first03On-Sensor Processingintegrated compute · sub-frame latency04Fast Data Pipelinehigh-throughput transmit · no bottleneckHuman GloveDEMONSTRATIONRobot SkinDEPLOYMENTINPUT: PHYSICAL CONTACTOUTPUT: ACTIONABLE TACTILE SIGNAL

Signal fidelity

Engineered for low noise from the contact surface up

Conventional flexible sensors typically trade fidelity for conformability. SoftShell is built around a signal architecture designed to preserve precision-task fidelity even as the sensing layer wraps real geometry.

TACTILE SIGNAL · 1 kHz CAPTUREconventional flex sensorSoftShell low-noise

Coverage

Built to cover surfaces, not points

Manipulation rarely happens at fingertips alone. SoftShell is designed for full-surface coverage across hands, palms and forearms — replacing point-contact instrumentation with a continuous tactile surface.

CONVENTIONALPOINT CONTACT ONLYSOFTSHELLFULL-SURFACE COVERAGE
Why same-sensor portability matters

Zero retargeting between human demonstration and robotic deployment

Most embodied AI pipelines hit a wall when human demonstration data has to be remapped onto robot hardware. SoftShell removes that translation step entirely. The sensor is the same, the data format is the same, and the embodiment is the same — across glove and skin.

Single sensing format

Identical signal layout from glove to skin — no per-platform calibration tax.

No retargeting

Demonstration data lands in deployment without remapping or interpolation.

Consistent embodiment

The data the model trained on is the data the robot perceives in production.

Unified evaluation

One sensing substrate makes A/B and longitudinal evaluation tractable.

Embodiment continuity

Glove

Worn on human hand

Demonstration capture

Pipeline

Same sensor format

Sensor → encode → transmit

Skin

Wrapped on robot end-effector

Production deployment

Applications

Built to enable the workloads tactile intelligence unlocks

SoftShell is designed to support the workloads that current robotic systems struggle to perform without rich, high-fidelity contact feedback.

A-01

Dexterous manipulation

Multi-finger grasping, in-hand reorientation and contact-rich tasks that require continuous tactile feedback rather than discrete fingertip pads.

A-02

Imitation learning

Capturing human demonstration with the same sensing format used at deployment, enabling cleaner imitation pipelines with no retargeting layer.

A-03

Embodied AI training

A standardized tactile substrate that makes contact-rich training data portable across platforms, models and research groups.

A-04

Prosthetic feedback systems

High-fidelity, low-noise contact signals at the prosthetic surface — designed to feed closed-loop feedback systems and BCI-adjacent stacks.

A-05

Robot retrofits

Bringing existing humanoid, industrial and research platforms online with tactile sensing without redesigning the hardware they ship today.

A-06

Tactile evaluation

A consistent substrate for measuring tactile capability and progress — across labs, platforms and product generations.

Engineering depth

Want a deeper technical walkthrough?

We share architecture details, signal characterization and integration paths under NDA with serious design partners and frontier labs.

SS / TACTILE-OSBUILT FOR EMBODIED INTELLIGENCE