01
See
Egocentric RGB, calibrated depth and spatial reconstruction establish what happened and where.
Teach robots throughhuman interaction.
Sapient Robotics turns multimodal human demonstrations into synchronized, physically grounded robot skills—preserving motion, contact and evidence from capture to simulation.
24 combined robot DoFSE(3) shared metric world4D contact correspondence
System
Video shows how a grasp looks.
Contact explains why it works.
MetaEgo—the platform behind Sapient Robotics—connects vision, articulated geometry and tactile evidence without confusing a model estimate for a physical measurement.
02
Track
Camera, hand and object motion are registered in one metric coordinate system.
03
Feel
Normal force, shear, contact and slip become synchronized physical evidence.
04
Transfer
Object-relative interaction is translated into feasible robot trajectories and tactile targets.
Pipeline
From a lived demonstration
to an executable interaction.
Each stage produces versioned artifacts with lineage, uncertainty and evaluation gates. Missing evidence remains unknown; simulated evidence remains simulated.
- 01
Capture
Wearable and external cameras record the human demonstration.
- 02
Synchronize
Independent devices are aligned on one auditable time axis.
- 03
Reconstruct
The world, articulated hand and moving object recover metric geometry.
- 04
Ground contact
Measured tactile evidence is attached to anatomical and object surfaces.
- 05
Simulate
Robot-specific contact and shear are evaluated in physics.
- 06
Transfer
The interaction becomes a training episode or executable robot target.
OBJECT_03 POSE ESTIMATED
4.82 N
REACHCONTACTGRASPLIFTROTATEPLACE
Contact Studio
Every signal,
on the same clock.
Inspect video, tactile force, hand pose, object pose and semantic phases as one synchronized episode. Correct model proposals without losing their provenance.
- World, camera, hand and object-relative views
- Anatomical tactile map and force history
- Measured, calibrated, derived and estimated layers
Research
Cross-embodiment transfer needs more than imitation.
Can persistent, skeleton-conditioned contact representations grounded by human tactile measurements transfer manipulation better than visual-only demonstrations?