Understanding contact.
Building intelligence.
CLAVIR is building the scientific foundation of Contact Intelligence.
Everything we learn — hardware designs, data, and results — is released open-source.
What is Contact Intelligence?
The ability to manipulate the physical world through rich, sustained contact. Building it is a single problem with three pillars.
Hands built to seek and sustain contact rather than avoid it — and to make what happens during that contact observable.
Data that captures sustained physical interaction, not just the collision-free trajectories between contacts.
Models whose representations and objectives are built around contact, instead of treating it as noise to be suppressed.
Robots are built to avoid contact.
Most systems treat physical contact as a risk to be minimized. That leaves the richest part of manipulation — sustained engagement with objects — the part robots handle worst.
Manipulation is contact.
Real manipulation is not a sequence of collision-free motions. Objects are held, pressed, rolled, and re-grasped. A system that cannot sustain contact cannot manipulate the physical world.
Foundations, not demonstrations.
Progress in contact needs shared instruments: hardware others can build, data others can use, and architectures others can test. CLAVIR builds all three as public infrastructure.
Contact Intelligence is the missing layer between seeing the world and changing it.
Every robot that works in the physical world eventually runs into contact, and that is where most of them fail. Vision can plan the approach, but the moment a hand closes, the outcome is decided by force, friction, compliance, and slip — properties no camera can measure, and that mainstream robot datasets rarely capture.
Force control itself is not new — impedance control and compliant assembly go back decades. What has never existed is contact at scale: hardware built to produce and expose it, datasets that record it, and architectures that treat it as signal rather than noise.
Everything downstream waits on that — assembly, tool use, in-hand manipulation, anything soft or deformable. A robot that cannot reason about contact is limited to the tasks that can be done without touching much.
An independent research lab building the foundations of Contact Intelligence.
We publish what we learn as open source so that progress on Contact Intelligence compounds across the whole field.
We pursue the fundamental research problems that unlock general-purpose physical intelligence.
Definitions, measurements, and instruments come before demonstrations. We build the layer the rest of the field stands on.
Hardware designs, data, and findings are published openly, so anyone can reproduce and build on them.
The people building Contact Intelligence.
A small group of researchers working across all three pillars, with backgrounds in robotics, physics, and mechanical engineering.
