The WELL (World Experience Learning Library) is a
multi-modal training dataset built from real-world
human physical activity. Motion capture, spatial
orientation, video, audio, and task context
collected by TRACE contributors during ordinary
daily work and leisure activities across diverse
environments and geographies.
The dataset is designed for training embodied AI
systems: humanoid robot control policies,
Vision-Language-Action models, and behavioral
foundation models that require grounded
understanding of how people actually move and
interact in unstructured settings.
It is offered under a license built to be the
easiest yes in your stack: generous, no-cost
carve-outs for research, development, and every
pre-revenue stage — and a de minimis fee (usually
under 1%) that begins only when your revenue does.
The WELL is released under the TRACE Human
Task Data License (HTD). The whole license
is built around one idea: stay out of your way until
the work is actually making money. You build freely
now, and share a little of the upside later — only
once there is real revenue, and even then barely.
Build freely. Download, train,
fine-tune, evaluate, benchmark, deploy, publish, and
even manufacture — at no cost. There is no fee to
negotiate, no gate to clear, and no permission to
wait on; you accept the license terms and get to
work. Academic research, experimentation, pilots,
demonstrations, and pre-revenue products all live on
the free side of the line.
Then a fee that barely registers.
Once a product is actually selling, or a service is
actually charging, the license fee is closer to a
rounding error than a cost — a sliver of the revenue
it rides on, sized to disappear into the margin.
Pre-revenue services get up to two years to grow
first, and the free-to-paid boundary is fixed in the
version you accept, so you always know exactly where
you stand.
A license that works with you.
Because TRACE earns only when you do, the whole
design points one way: getting the data into your
systems and keeping it working — for you and for the
people who produced it.
Clean to build a company on. The
fee is a cost of goods, not equity — it never touches
your cap table or takes a stake in what you build.
And properly licensed, KYC-sourced,
provenance-tracked training data is the kind of clean
sourcing that holds up in diligence.
Provenance and verification. Every
record carries provenance metadata, and the dataset
is built for downstream traceability through
provenance records and watermark verification as
those methods mature. This protects honest licensees
and keeps the contributor revenue-share accounting
auditable.
Never used to identify people. The
license categorically prohibits using the data — or
any model trained on it — to identify, re-identify,
or biometrically recognize the people who produced
it. This is a core term, not a configurable
setting.
The full license suite — base license, standard
commercial terms, and policies — is available for
review on request.
TRACE is currently in private alpha. The WELL
dataset is under active development as our first
contributors begin collecting verified task
data.
We are building toward commercial licensing of the
dataset for research and product development in
embodied AI. If you are a researcher, founder,
robotics company, or AI lab with interest in early
access,
partnership, or licensing, we would like to hear
from you.
What we can discuss now:
- Data modalities, formats, and collection methodology
- Licensing structure and terms
- Research collaboration and early access programs
- Custom data collection for specific task verticals
For research inquiries and data licensing:
@exosequitur on Telegram
For general questions about TRACE, join our
community:
TRACE Telegram group
|
@TRCDynamics on X
In private alpha - public beta coming soon
Check the FAQ