The DreamVu Pipeline
We customize the pipeline for every data challenge.
Each stage uses the best method available — open source where open source is better, our own where it is not. We publish in this field, so we track what changes, and we swap components when something better comes out. You tell us what your model needs to learn, and we build the pipeline for it.
Stage 01
Capture
Exocentric
Alia, our own camera. A single-shot 360° stereo panorama with depth, from one camera position.
Egocentric
A head-mounted camera on the worker, synchronized to the Alia stream frame for frame. Some datasets, such as navigation, use the ego camera on its own.
Wrist
An optional third camera for close manipulation and grasp detail.
Other cameras
The exocentric stream is Alia. Ego and wrist cameras are set by what the program needs, including your own.
Environments
Working venues. We have access to hundreds of venues across a range of industry segments.
Stage 02
Annotate
Method
AI does the first pass. A person reviews every batch. Segmentation, tracking, action boundaries, and task breakdown are set per program.
Components
The best current model for each task, replaced as the field moves. Our own components where published methods are not good enough.
Stage 03
Deliver
| Output | Capture | Used for |
|---|---|---|
| World model data | Alia | Generative video and world model training |
| VLM datasets | Ego + exo | Vision-language model fine-tuning |
| Navigation datasets | Ego only | Navigation and traversal policy training |
| Loco-manipulation datasets | Ego + exo + wrist | Loco-manipulation and action policy training |
| Simulation environments | Alia capture | Any USD-compatible simulator — in development |
Standards
OpenUSD. LeRobot (RLDS). Open X-Embodiment. VLM, navigation, and panorama deliveries have no single standard, so we agree the schema with you before capture starts.
If you need something that is not on this list
Tell us and we will scope it. We also run USD conversion with physics and domain-randomized rendering from our own capture.
Quality
Four gatesEvery capture passes four gates before we deliver it.
G1 Calibration
Checked for every rig, every session.
G2 Alignment
All camera streams confirmed in sync before annotation starts.
G3 Annotation review
A person reviews every batch.
G4 Rejection
Batches that fall below the threshold are captured again.
Full numeric specification available under NDA
Tell us what your model needs to learn.
Capture programs, research collaboration, and dataset partnerships.