Work is contextual.
The same movement means something different in every environment.
FIG.01 · PHYSICAL INTELLIGENCE UNIT
The data infrastructure layer for Physical AI.
Signal inventory0 / 6
01 Egocentric visionRDY
02 Human actionRDY
03 Machine stateRDY
04 Workflow contextRDY
05 OutcomeRDY
06 Rights & provenanceRDY
DrawingDS-001.PNG
UnitDataset
TitleReality → Signal
DWG no.DS-001 · REV A
Scale1:4 · ISO 30°
Sheet1 of 1
( 00 )Document index
DS-001 · Operational data architecture
( 01 )The premise · DS-002
Physical intelligence cannot learn from pixels alone.
Every shift contains the intelligence machines are missing.
Dataset is the data infrastructure layer for Physical AI capturing authentic industrial work and transforming it into rights-cleared, operational-ready data.
( 02 )The constraint
3 gaps · observedIndustrial data exists. Buyer-ready datasets do not.
Reality is multimodal.
Vision, sensors, systems and outcomes must arrive together.
Trust is structural.
Rights and provenance are part of the data not paperwork after it.
( 03 )The Dataset system
Reality,
made learnable.
Infrastructure that carries physical AI from raw work to model-ready intelligence.
01Dataset CaptureObserve the work
02Dataset OSStructure the episode
03DataFaceRelease the intelligence
( 04 )Signal vocabulary · QIE
Reality,
made learnable.
A Qualified Industrial Episode synchronizes what happened, where it happened, and what came next.
01RGB · DEPTH
Egocentric vision
Live02POSE · TRAJECTORY
Human action
Live03PLC · TELEMETRY
Machine state
Live04TASK · ENVIRONMENT
Workflow context
Live05SUCCESS · FAILURE
Outcome
Live06VERIFIED · TRACEABLE
Rights & provenance
Live
( 05 )Research program
ACTIVE THESISWorld Action Models · WAM-01
Can machines
inherit experience?
We are exploring models grounded in the states, actions, failures, recoveries and outcomes of real work.
Controlled research program( 06 ) Request accessDS-001 · OPEN
Build from the world as it is.
For frontier AI teams and industrial partners shaping what comes next.
Talk to us →