Stow

Building the data layer for Physical AI.
Real-world human demonstrations → high-quality training data → smarter robots.

[Stow @StowData]

A. Basic – timestamps and task segmentation
B. Rich – objects, surfaces, tools and environment
C. Robotics-ready – hand poses and action labels
D. Not sure yet – tell us what you're training

Which level would you choose? Backed by @LiftoWorld, our fundraising partner.

[Stow @StowData]

Five questions robotics labs ask first:

Can I filter by capture setup?
Do I get refused clips?
Which formats are supported?
How fast is a quote?
What about consent?

It needs clear records, usable formats and transparent decisions. Backed by @LiftoWorld, our fundraising partner.

[Stow @StowData]

Before any payout moves, two things must be decided:

A signer is wired.
A per-task spend ceiling is set.
Human approval comes first. Automation only moves within defined limits.

Built for accountability, not blind automation. Backed by @LiftoWorld, our fundraising partner.

[Stow @StowData]

Six skills on the robot learning roadmap:

Sequencing
Object destination
Planning
Deformables
Categorisation
Tool use
Each skill builds on the previous one, turning everyday household chores into learnable robot capabilities.

Backed by @LiftoWorld, our fundraising partner.

[Stow @StowData]

The next generation of kitchen robots will learn from:

Ordinary people
Everyday tasks
Real demonstrations
Recorded lessons
But who taught the robot, and was that lesson any good?

STOW keeps that record. Backed by @LiftoWorld, our fundraising partner.

#RobotLearning #Robotics

[Stow @StowData]

Three steps before payouts move:

Signer is wired
Per-task spend ceiling set
Manual approval first

Automation needs clear limits and human control. No spending happens until both decisions are made. Backed by @LiftoWorld, our fundraising partner.

#Robotics #AI #Trust

[Stow @StowData]

Four principles of data provenance:

No intermediaries
No scraped provenance
Consent that names AI training
Nothing to reconstruct later

Every dataset should have a clear connection to its contributor. Public availability is not permission. Backed by @LiftoWorld, our fundraiser.

[Stow @StowData]

A rejection should explain three things:

Which rule failed?
Whose words explain it?
What measurement overruled the model?
A simple "Rejected" isn't enough. Robot training needs transparent decisions that contributors can actually understand. Backed by @LiftoWorld, our fundraiser

[Stow @StowData]

Three possible outcomes for robot training data:

Accepted → All checks passed
Refused → A rule was missed
Already held → Too similar to existing data

Every decision has a reason. Better data needs clear evaluation, not just collection. Backed by @LiftoWorld, our fundraiser

[Stow @StowData]

How does a robot learn to fold?

Watch a person
Observe the movement
Learn the sequence
Evaluate the result
Improve through feedback

The real challenge is turning human demonstrations into reliable robot actions. Backed by @LiftoWorld, our fundraising partner.

#RobotLearning

[Stow @StowData]

Four things off the table:

Plates → Sink
Glasses → Sink
Scraps → Bin
Leftovers → Cupboard

Sounds simple, right? A robot needs to recognise, categorise and place every item correctly. One wrong decision can ruin the whole task. Backed by @LiftoWorld, our fundraising partner.

[Stow @StowData]

Five chores robots still struggle with:

Clearing the table
Changing a bin bag
Emptying a dish rack
Sweeping into a dustpan
Tidying a cluttered surface

STOW is working towards training robots for the real world. Backed by @LiftoWorld, our fundraising partner.
#Robotics #AI

[Stow @StowData]

Teaching a robot to clean up:

Framing
Mess visible
Containment
Completeness
Continuity
Duration
Visibility
One subject

Even a simple task needs rules. Robots learn by observing, gathering, and getting the job done. Backed by @LiftoWorld, our fundraising partner.
#Robotics

[Stow @StowData]

New project analysis is live on Lifto.

STOW turns everyday household tasks into valuable robot training data, with transparent AI-powered verification and contributor rewards.

Explore STOW 👇
https://t.co/cRDxRe3zHE

#Robotics #AI

[Lifto @LiftoWorld]

📰 TODAY'S PROJECT ANALYSIS BY LIFTO

Project: @StowData
Category: Robotics / Robot training data

STOW turns a 20-second household cleanup filmed on your phone into robot training data. An AI agent checks every clip against 8 written rules. Accepted clips get paid in USDC, and refused clips get the exact reason why.

🔎 THE VERDICT
✅ Founder with 8 years in robotics, taught at Carnegie Mellon
✅ Every accept/refuse decision published on a public ledger
✅ 35% of the raise goes straight to contributor payouts

Check out STOW on Lifto 👉 https://t.co/p6nqTsEhK5

New project analysis every day. Stay tuned 🔔

[Stow @StowData]

Eight rules for better robot demos:

Framing
Mess visible
Containment
Completeness
Continuity
Duration
Visibility
One subject

A good demo isn't just about what the robot can do. It's about showing the entire process, clearly and honestly Backed by @LiftoWorld,fundraising partner

[Stow @StowData]

How honest robot quoting works:

Order in
Check supply
Count actual stock
No quote
Explain why

A quote should reflect what you can actually deliver, not what you hope is available. Honest data makes better decisions. Backed by @LiftoWorld, our fundraising partner.
#Robotics

[Stow @StowData]

Lab: “500 clips at grade 0.80+?”
Us: “One clip clears 0.80 right now.”
Lab: “Can you quote anyway?”
Us: “No. We don’t quote against supply we don’t hold.”

That’s how datasets don’t get sold twice. (Dramatised; the order details are from the whitepaper.) https://t.co/wY5KtpXGaH

[Stow @StowData]

Robot-speak, translated

Order of operations → sequencing: what to gather before what.
Semantic placement → object sense: tool, rubbish, or back in the cupboard.
Failure data → recovery: what to do when it smears.
Terminal state → judgement: when it’s actually clean.

[Stow @StowData]

Four design principles behind our decisions:

The model answers, the code decides
Some rules score instead of gate
Measured facts beat the model
A broken agent never costs the contributor the work

https://t.co/wY5KtpXGaH

[Stow @StowData]

Today’s mess forecast ⛅

Morning: scattered rice. Noon: tea, with a chance of spill. Afternoon: crumbs and debris. Evening: cluttered surface.

Outlook: perfect filming weather. Phone on a stand, both hands in frame. https://t.co/wY5KtpXGaH