01/Automotive
Autonomous vehicles & ADAS
Multi-sensor perception and validation data for ADAS and autonomous driving programs.
Verified data, on demand.
License verified datasets for autonomous vehicles, robots, and industry, or contribute your own sensor data and earn from every license. One neutral marketplace between the physical world and the models that run on it.
Domains
Verified data for every Physical AI domain.
Every dataset is verified and ready to train. Browse by domain, or stream straight into training.
02/Robotics
Robotics & mobile autonomy
Manipulation, navigation, and tracking data for warehouse and mobile robots.
03/Embodied AI
Humanoids & embodied AI
Real-world task footage for humanoids and general-purpose robots learning to act.
04/Energy
Energy & digital infrastructure
Aerial inspection and digital-twin capture for infrastructure and energy assets.
The Switzerland of Physical AI data.
Deepen Refinery is infrastructure, not a competitor. We don't build models. We don't keep your data. We don't train competing systems. Every dataset is verified, with full provenance.
01/✓
No model building
We refine and verify data. We never build models that compete with our customers.
02/✓
No data retention
Your data moves through the pipeline. We retain no rights to use it elsewhere.
03/✓
Vendor neutral
Every foundation-model lab can trust Deepen as an independent supplier.
04/✓
Full provenance
An immutable chain from sensor to delivery, auditable end to end.
Contribute and earn. License and train.
One neutral marketplace, two sides, with no conflict of interest between them.
Data Partners
Contribute and earn.
Fleet operators, OEMs, and research labs. Contribute sensor data, we refine and verify it, and you earn from every license. Partner identity stays protected until NDA.
up to 80%
revenue share · 0% listing fee · Net-30 payouts
ML Teams & Labs
License and train.
ML engineers, foundation-model labs, AV and robotics teams. License verified datasets, compose custom packages, and stream straight into training via API.
LeRobotMCAPROS2HDF5nuScenes
Enter at any stage. The data is refined, never resold as-is.
Five capability stages. Engage with all of them, or just the one you need.
01
Collect
Partners contribute raw sensor data under a signed agreement.
02
Curate
Deduplicate, balance, score diversity, and surface edge cases.
03
Enrich
Calibration verification, 3D annotation, sensor fusion, PII removal.
04
Validate
Safety Pool scoring, scenario coverage, and audit-trail generation.
05
Deliver
Native export and API stream with an immutable provenance chain.
Stream verified data into your pipeline.
Three lines to authenticated access. Stream certified frames straight into your training loop via REST or WebSocket. No download required.
# install
pip install deepen-refinery
# authenticate and stream certified data
from deepen import RefineryClient
client = RefineryClient(api_key="dr_live_ck_...")
for frame in client.stream("DRX-0481", format="lerobot"):
model.train_step(frame) # verified ✓Built for the training loop.
- → REST and WebSocket streaming, no migration
- → Native: LeRobot, MCAP, ROS2, HDF5, nuScenes
- → Calibration metadata preserved on every frame
- → Three key types: consumer, partner, processor
Python SDKREST APIWebSocketLeRobot
Verified to the standards Physical AI is held toASAM OpenLABELSOC 2 Type IIISO 27001TISAXEU AI Act-readyEuro NCAP 2026Safety Pool
How the marketplace works.
What is the Refinery?+
A neutral marketplace for Physical AI data. License verified datasets for autonomous vehicles, robots, and industry, or contribute your own sensor data and earn from every license.
How does revenue share work for partners?+
Contribute verified sensor data and earn up to 80% of every license, with a 0% listing fee and Net-30 payouts. Your partner identity stays protected until an NDA is signed.
How do I access licensed data?+
Stream certified frames straight into your training loop over REST or WebSocket in LeRobot, MCAP, ROS2, HDF5, nuScenes, or native formats. No download required.
Is my data used to train competing models?+
No. We build no models, retain no rights to your data, and stay vendor neutral, with full provenance from sensor to delivery on every dataset.
Which domains does the catalog cover?+
Autonomous vehicles and ADAS, robotics and mobile autonomy, humanoids and embodied AI, and energy and digital infrastructure.
Train on verified data.
License verified Physical AI data, or contribute yours and earn, on the one neutral marketplace both sides can trust.