Technology

Three technologies
behind a hand that grips.

Touch, to read contact at the fingertips. Force control, to decide how hard to press based on what it reads. And learning, which gets more precise the more it's used on site. Together, these three let our hand handle objects of any shape or firmness—without dropping or crushing them.

01 Sensing a slip before it happens.

Pressure sensors packed into a 3 mm square at the fingertip catch the subtle pressure shift that appears just before an object slips—and turn it into a cue to re-grip before anything falls.

368 points / 1 kHz

Sensing points / Sampling rate

02 Moving by force, not position.

The target isn't “how far to move” but “how hard to press.” Soft things stay uncrushed; hard things stay securely held.

1 ms / 8 ms

Command cycle / Response latency (avg.)

03 The more it works, the more it fits the job.

The model updates gradually with contact data gathered during operation, steadily converging on the right touch for the objects handled on that site.

2–4 weeks after deployment

Typical time to adapt to a site

HOW IT WORKS

One full loop
every millisecond.

A camera locates the object, the fingertips read the contact, the controller decides the force, and the hand moves. Split those four steps across separate devices and every handoff adds delay. AWAI's hand runs all four on a single board.

Touch

1 / 3

Sensing a slip before it happens.

We've packed 368 pressure-sensing elements into a 3 mm square at the fingertip. Just before an object starts to slip, the pressure across the contact surface shifts ever so slightly. We read that shift at 1 kHz and turn it into a cue to re-grip before anything falls.

A hand touching snow, evoking the feel of something soft that changes shape.

1.1 How Detection Works

Sensing elements 0.15 mm apart feel pressure shift toward the center as a slip begins.

Milliseconds before the object moves, the grip tightens just enough—so soft items keep their shape.

Touch

1 / 3

Sensing a slip before it happens.

1.1 How Detection Works

Sensing elements 0.15 mm apart feel pressure shift toward the center as a slip begins.

Milliseconds before the object moves, the grip tightens just enough—so soft items keep their shape.

Touch

1 / 3

Sensing a slip before it happens.

1.1 How Detection Works

Sensing elements 0.15 mm apart feel pressure shift toward the center as a slip begins.

Milliseconds before the object moves, the grip tightens just enough—so soft items keep their shape.

Force Control

2 / 3

Moving by force, not position.

Most robots aim for “how far to move.” AWAI's hand aims for “how hard to press.” Soft things stay uncrushed; hard things stay securely held. It decides the right force for each object, on the spot.

A robotic hand gently pinching the stem of a single red gerbera.

2.1 Our Approach to Control

Force-based control holds objects with the intended force, even if size or position is slightly off.

Each joint gets a command every 1 ms, with an 8 ms average delay from sensor to motor.

Force Control

2 / 3

Moving by force, not position.

2.1 Our Approach to Control

Force-based control holds objects with the intended force, even if size or position is slightly off.

Each joint gets a command every 1 ms, with an 8 ms average delay from sensor to motor.

Force Control

2 / 3

Moving by force, not position.

2.1 Our Approach to Control

Force-based control holds objects with the intended force, even if size or position is slightly off.

Each joint gets a command every 1 ms, with an 8 ms average delay from sensor to motor.

On-Site Learning

3 / 3

The more it works, the more it fits the job.

Even for the same task, the objects and their placement differ from site to site. The model updates gradually with contact data gathered during operation, and within a few weeks of deployment it converges on the right touch for the objects on that site.

An engineer working on a laptop in front of test equipment lined with cables and speakers.

3.1 How Learning Works

Every grip, including re-grips, becomes training data, and the model is updated overnight.

After a safety check, updates go live gradually; within 2–4 weeks, re-grips become rare.

On-Site Learning

3 / 3

The more it works, the more it fits the job.

3.1 How Learning Works

Every grip, including re-grips, becomes training data, and the model is updated overnight.

After a safety check, updates go live gradually; within 2–4 weeks, re-grips become rare.

On-Site Learning

3 / 3

The more it works, the more it fits the job.

3.1 How Learning Works

Every grip, including re-grips, becomes training data, and the model is updated overnight.

After a safety check, updates go live gradually; within 2–4 weeks, re-grips become rare.

SPECIFICATIONS

Key Specifications

Item

Value

Condition

Pressure-sensing elements

368 per finger

Per 3 mm square

Sampling rate

1kHz

All elements simultaneously

Minimum detectable pressure difference

0.8kPa

At 25 °C room temperature

Item

Value

Condition

Pressure-sensing elements

368 per finger

Per 3 mm square

Sampling rate

1kHz

All elements simultaneously

Minimum detectable pressure difference

0.8kPa

At 25 °C room temperature

Item

Value

Condition

Pressure-sensing elements

368 per finger

Per 3 mm square

Sampling rate

1kHz

All elements simultaneously

Minimum detectable pressure difference

0.8kPa

At 25 °C room temperature

PUBLICATIONS

Publications & Patents

2026

Stabilizing Grasps Through High-Speed Slip Detection / International Robotics Conference

2025

Manufacturing Method for Distributed Fingertip Pressure Sensors / Patent Application

2025

On-Site Adaptation of Grip Force Models Using Contact Data / Domestic Conference

2026

Stabilizing Grasps Through High-Speed Slip Detection / International Robotics Conference

2025

Manufacturing Method for Distributed Fingertip Pressure Sensors / Patent Application

2025

On-Site Adaptation of Grip Force Models Using Contact Data / Domestic Conference

2026

Stabilizing Grasps Through High-Speed Slip Detection / International Robotics Conference

2025

Manufacturing Method for Distributed Fingertip Pressure Sensors / Patent Application

2025

On-Site Adaptation of Grip Force Models Using Contact Data / Domestic Conference

CONTACT

Want to try it
on your own line?

Tell us what you handle and how, and we'll set up a live demo. You can also send us your own parts or products so we can confirm in advance that we can grip them. Whether you're close to a decision or just have questions about the technology, we'd love to hear from you.

Talk to an Engineer

View Open Roles

Technology

Technology

Technology

TECHNOLOGY

Three technologiesbehind a hand that grips.

01

Sensing a slip before it happens.

02

Moving by force, not position.

03

The more it works, the more it fits the job.

HOW IT WORKS

One full loopevery millisecond.

Sensing a slip before it happens.

Sensing a slip before it happens.

Sensing a slip before it happens.

Moving by force, not position.

Moving by force, not position.

Moving by force, not position.

The more it works, the more it fits the job.

The more it works, the more it fits the job.

The more it works, the more it fits the job.

SPECIFICATIONS

Key Specifications

PUBLICATIONS

Publications & Patents

CONTACT

Want to try iton your own line?

Three technologies
behind a hand that grips.

One full loop
every millisecond.

Want to try it
on your own line?