Precise force data, from arm to fingertip.

Wrist force and torque, fingertip touch, on one timeline.

Book a demo

Real footage, teleoperated. Plug inserted, lamp on.

Wrist force

1.20N

Six-axis force and torqueVisualized

Axes

Fx: +0.0
Fy: +0.0
Fz: +0.0
Tx: +0.00
Ty: +0.00
Tz: +0.00

Most of what robots still can't do is contact.

Pushing, fastening, pressing into place. That work runs on force and touch, not on sight.

We record the force skilled workers actually feel, from the arm to the fingertip, and teach it to robots.

Real footage of a robot hand about to insert a plug into a power strip. — What a camera records.

Position goes to the robot. Force comes back to the hand.

01The operator's hand moves. The robot follows.
02The robot's fingertip meets the plug. Force appears.
03That force returns to the glove. The hand feels resistance.
04Both directions are recorded on one timeline.

Operator

Robot, teleoperated

Push it in.

Drag the plug into the socket. It gets heavier as it goes in.

Stage: Approach
Wrist force: 0.0 N
Depth: 0 mm

Visualized

Capture it, shape it, teach it.

Acquire

Dream Catcher

An arm that teaches by feel, a glove that records touch, a studio for capture and review, and Lucid for bilateral control.

Live demo available

Explore Dream Catcher

An operator's hand in a master glove, next to the robot arm.

Dataset

Dream Set

Nine layers per episode, from task video to wrist torque, recorded on one clock.

Available

Explore Dream Set

Model

Morpheus

A contact intelligence foundation model, built on open backbones with a layer that understands force.

Coming soon

Force in, action out.

Before we talk haptics, we start with contact data.

83%

Success when the policy learned force and torque. Video alone reached 22%.

ManipForce, 2025. Six tasks, averaged, rounded.

+23.2pp

Gain from adding force-aware fusion to π0, from 37.3% to 60.5%.

ForceVLA, 2025. π0 baseline, five tasks.

1.09s

To mate a moving HDMI socket. A person takes 1.04 s.

SHIELD, RSS 2021. Dedicated force and vision system.

One insertion. Nine layers of data.

Real footage of the moment of insertion.

Task view

Wrist view

Joint and hand pose

Force and torque

Fingertip touch

Operator command

Force and stiffness

Gaze

Task stage

Time

0.0 s1.02.03.0 s

Kinesthetic layers. Shapes are illustrative.

Every task has its own force signature.

Driving a screw

Sustained torque

Angle grinding

Vibration load

Plug insertion

Insertion peak

Pouring

Shifting mass

Precision finishing

Fine pressure

Electronics, automotive, shipbuilding, heavy industry.

Research and notes.

View all

Data acquisition

Stable contact teaching: managing energy in bilateral control

Data acquisition

Firm surfaces. Natural movement.

Data acquisition

Bimanual haptics that let human hands lead

Send us a contact task.

We will answer with a demonstration.

Contact Request a data sample

Help teach robots the sense of touch.

See openings