Capture
Capture real-world demonstrations with synchronized motion, video, and structured task data. Start from real signal: task context, traces, and motion that can be used again.
Phi9 — physical AI lab
Physical AI research, data systems, and deployable intelligence.
Phi9 captures real-world behavior, structures it for training, and evaluates what actually transfers — so research can become usable capability.
The loop
From capture to evaluation, the physical AI loop.
One loop. Four stages. The work is turning real-world behavior into signal, training against it, and measuring what survives contact with reality.
Multiply
Multiply scarce data through retargeting, simulation, augmentation, and better structure. Stretch each capture further without letting intent or task definition drift away.
Train
Train policies and research systems on data that preserves intent, motion, and context. The point is not isolated models; it is a training layer that stays close to reality.
Evaluate
Evaluate what generalizes through benchmarks, failure analysis, and transfer tests. Treat deployment feedback and failure traces as part of the same loop.
-
4
Loop stages
-
12+
Modalities per capture
-
25+
Environment types
Featured system
The capture system behind the loop.
The Phi9 MoCap Rig records full-body motion and first-person video together, timed and labeled by task. The output is reusable training data instead of one-off recordings, so the same demonstration can move through training, fine-tuning, and deployment.
See the Data Research questions
The bottlenecks we are actively working through.
These are not abstract themes. They are the constraints shaping the systems, experiments, and artifacts we are building now.
- 01
Data that carries intent, not just observation.
Most pipelines record visible motion but lose the task underneath it. We are working on capture that preserves action, context, and what the body was trying to achieve.
- 02
Physical data is expensive.
You cannot scrape physical behavior. Every demonstration needs a rig, a subject, a calibration, and a clean task. The work is making each capture travel further without losing signal.
- 03
Benchmarks that predict real-world performance.
A benchmark score means little if a policy falls apart on an unscripted task. We care about evaluation that predicts transfer, failure, and what survives outside the benchmark.
- 04
One loop, not three stages.
Capture, training, and evaluation still get treated as separate departments. We are trying to wire them into one visible loop so progress does not disappear between stages.
Research
Projects, notes, and technical progress from the lab.
Open work from the loop: systems, experiments, and technical questions published as they evolve.
Methods
The concrete systems we are building around the loop.
Methods should feel like work, not philosophy. These are the concrete layers we are building now so demonstration, training, and deployment stay connected.
Capture surfaces
Rigs, sync, task framing, and sensor traces that start with the real world instead of a benchmark-only abstraction.
Data structure
Labels, schemas, exports, and task boundaries that keep demonstrations reusable across research, training, and downstream tooling.
Training systems
Retargeting, augmentation, and policy pipelines that stretch scarce behavior while preserving intent, motion, and context.
Evaluation layer
Benchmark fragments, transfer tests, and failure traces that keep the loop honest about what actually generalizes.
Contact
Building in physical AI?
If you are working on data, research systems, or deployment infrastructure for physical intelligence, write to us.
Write to
Read the manifesto