The end of a robot assembly line, empty
STATION 01 · CONCEPTBuilt

DropFly Robotics

Built.

A humanoid comes off a line the way a car does. Torque checked, joints checked, power on, boxed. Every machine on that line passed the tests written by the people who built it.

The ask

Somebody should have tried to break it first.

$0 raised of $25,000 goal
No deadline · Non-dilutive · Open until we get there
Back this raise

You are not buying a piece of the company. No equity, no shares, no payout if this works. DropFly is a for-profit company, so it is not tax-deductible either. We keep it that way on purpose: we are asking robot companies to trust us to be neutral, and that is easier when we do not owe investors a return.

On 7 September 2026, XPeng’s IRON humanoid walked off a newly commissioned production line in Guangzhou under its own power, on a line where more than 80% of core processes are automated. Mass production is slated for this year, with the first units going to the company’s own stores and campuses and commercial delivery in 2027. None of the coverage records an independent safety evaluation anywhere in that sequence. Electrek, 7 Sep 2026 ↗

The problem

You wouldn't let a restaurant write its own health inspection.

Cars have crash tests run by people who don't work for the car company. Medicines have trials someone outside the drug company signs off on. Restaurants get a health inspector who can shut the kitchen down. Buildings have inspectors. Planes have inspectors.

In every one of those, somebody outside the building is allowed to look, and allowed to say no.

Robots don't have that yet. A company builds a robot, tests it itself, decides it's ready, and ships it into a warehouse where people work. There's no outside inspector, because the job doesn't exist. Nobody has built it.

That's not a scandal. It's just early. Somebody has to be first, and we'd like it to be us.

What we do

Three jobs, in plain terms.

01

We test robots we didn't build.

We're not making our own robot. That's the point, we have nothing to gain from any particular robot passing. Our whole value is that we're not on anyone's team.

02

We check it was allowed to do that.

Not just "did it work" but "was it supposed to?" Every action the machine takes gets checked against what it actually had permission to do, and written down so somebody can look later.

03

We try to break it first.

Anything works in a tidy lab. We run the awkward situations, the ones a real warehouse or a real hallway will produce, so the failures happen with us instead of with somebody's employee.

What a test looks like

Two people, a robot, and a room built to trip it up.

A test starts in an ordinary room. The robot is given a job. Two of us watch what it does and write down what happened, not just whether it finished, but whether it was ever allowed to do what it did.

Concept A test bench of the kind this raise pays for. We don't have this room yet, that's the point of the raise.

Then we look through the robot's eyes. Before it ever moves in a real room, it runs the same job hundreds of times inside a simulated one, a room we built specifically to give it trouble. This is roughly what it sees: shapes it has decided are objects, boxed and tracked, and a floor it thinks it understands.

Concept The machine's-eye view of a test environment. Every box is something the robot has decided is a thing. When it boxes the wrong thing, that's a finding.

And then the part nobody else publishes. Below is a real run out of our own harness, ten trials of one scenario. The quick automatic check called every single one a failure. A panel of three independent judges looked at the same ten and refused to sign four of them.

On four of ten trials, the machine and the judges disagreed.

Quick check 10 fail
Panel of three 6 fail · 4 not sure
✕ failed — not sure 4 of 10 disagreed

Honest caveat, because it matters more than the chart: the thing being tested here is a stand-in, not a real robot brain. The harness is real and these numbers are real. A real robot brain needs a machine we don't own yet, which is what the $25,000 buys.

Where the money goes

$25,000, and we'll tell you exactly what it buys.

Not a lab. We're not going to pretend otherwise, a small industrial unit in Los Angeles runs $2,500–3,000 a month before you've built anything in it, so a year of rent alone would eat this whole raise. We haven't earned a lease yet.

What it buys is a bench that can actually test the robot brains companies are shipping today, rather than toy versions of them. That runs in a garage, and it's enough to produce results nobody else is producing.

A computer big enough to run the real thing $16,000 Today's robot-control models are enormous. On a normal computer you can only run the shrunk-down versions, which tells you nothing about the one that actually ships. This is the machine that runs the real one.
A pair of robot arms $4,500 One arm a person moves by hand, one that copies it. That's how you record a human doing a task correctly, then make the software attempt the same thing and see where it goes wrong.
Cameras, mounts, bench and guarding $2,000 Every trial filmed from several angles, because "it failed" is worthless without being able to show what happened. Plus a bench solid enough to bolt an arm to and the guarding to run it safely near people.
Card fees and sales tax $2,500 Counted here out loud rather than quietly coming out of the equipment budget. Card processing takes a cut, and California charges tax on the hardware.
Total$25,000

No salaries. No rent. No software subscriptions. Every dollar is equipment that still exists after the raise is over.

Who we are

Two founders, a garage, and a job nobody has taken.

Rio Allen
Co-founder & CEO

The direction, the engineering, the test rigs, and the harness that produced the record at the top of this page.

Erik Scott
Co-founder & COO

The operations, the partnerships, and making the case that an independent standard should exist at all.

We're two Black founders building the referee for an industry that has almost none of us in it. If this exists, you helped make it exist, and that's a thing you'll be able to point at.

The deal

You're not buying a piece of the company.

What this is

  • A donation. You give money, we buy equipment, we tell you what we found with it.
  • Non-dilutive. That's the industry word for "nobody gives up ownership." We keep the company; you don't take a piece of it.
  • A deliberate choice. We're asking robot companies to trust us to be neutral. That's easier when we don't owe investors a return.

What it isn't

  • Not shares. No equity, no stake, no stock certificate, no payout if this works.
  • Not tax-deductible. DropFly is a for-profit company, not a charity. We're not going to dress it up as one.
  • Not a pre-order. The merch below is a thank-you, not a product you're buying.
The ledger

Every contribution, as it lands.

Anonymous by design, amounts and timing only. We never publish who backed us.

Loading…
Pulling the latest contributions.

Last 10 contributions · amounts and reward tier only · no names, emails or locations are collected or shown here.

Thank-yous

Every tier gets you something to wear.

Made through MerchFly, our own merch studio real blanks, printed properly. The amounts are minimums, not prices. If you'd rather just back the work and skip the merch, name your own amount from $1 up.

Any amount

Just back it

No merch, no shipping, nothing to size. From $1 up.

Give any amount
$100+

Structured cap

Six-panel cap, adjustable strap, mark on the front panel. One size.

Back at $100+
$250+

Heavyweight tee

Heavyweight cotton, boxy premium fit, left-chest mark. S–XXL.

Back at $250+
$500+

Premium hoodie

Heavyweight brushed fleece, streetwear cut, left-chest mark. S–XXL.

Back at $500+

Size chosen at checkout · ships once we place the order · every backer gets the lab updates.