We built this demo in collaboration with a clinical pathology lab. It shows a single system doing real lab tasks: tool use, precision manipulation, and high level planning.
This was done in one take with no teleop and uses our skills based AI to enable generality while staying deterministic and reliable for real world workflows.
First look at our n house tactile sensor prototype. It detects taps, motion, and multiple contacts with force magnitudes. Built to be cheap, robust, and replaceable. Not used in our demos yet, still a work in progress.
At a high level we pair simple tactile sensing with torque transparent joints to recover rich contact without expensive sensors.
Here we show assembly of parts inspired by things that our founder designed at SpaceX. In one continuous sequence it handles pick and place, insertion, nut threading, and in-hand manipulation.
For low volume parts a dedicated production line isn’t viable so our general-purpose system can drop into existing workflows and can be quickly repurposed.
This shows our hand rotating a large nut on a bolt at very high speed, fully in real time. It works because the hand is mechanically backdrivable and torque-transparent, allowing it to naturally comply with the nut geometry.
The point isn’t raw speed, it’s reliability and capability. By letting the hardware handle variability mechanically we keep the control and software stack simple and robust.
This shows a finger stopping on contact with a feather using only the physics of our direct drive actuation.
Backdrivability and torque transparency let it sense contact directly through its drive currents, enabling robust, compliant interaction and clean learning signals.
This first public demo shows our approach: fully backdrivable, mechanically compliant hardware that naturally adapts to the world, enables safe interaction, and is built for real deployment and large-scale learning.
We went on a podcast with some great experts in the field and it was a pleasure to discuss hand design with them!