About Fireloop
Software learned to test itself. Robots haven't yet.
Twenty-five years ago software shipped on the developer's word. Then testing became a discipline, then a profession, then an independent function nobody would ship without. Physical AI is at the start of that curve: robots are cleared for a warehouse floor on a demo, a pilot and the vendor's assurance. Fireloop exists to give physical AI what software already has: an independent, repeatable way to prove a machine is ready before people share a floor with it.
The pattern
Every technology that touched people ended up with an independent test.
Cars got crash tests. Aircraft got certification. Software got QA, then continuous testing, then an industry of tools whose only job is to find where the code fails. In each case the maker's own word stopped being enough once the stakes were high and the volume was large. Robots working beside people have reached both.
Ship on the developer's word
The team that wrote it decides it works. Failures are found by customers.
Test as a discipline
Dedicated testers, written cases, measured coverage. Failures are found before release, but only the ones somebody thought to write.
Independent, automated, continuous
Tooling generates and runs the cases, every build, with evidence a third party can read. Nobody ships without it.
A demo, a pilot and a promise
Each vendor tests their own robot, in their own way, on their own schedule. The floor finds the rest. This is where Fireloop comes in.
What independent means
We build neither the robot nor its software. We build the software that tests it.
A verdict from the robot's maker is a claim. A verdict from a pilot is an anecdote. A verdict produced by software with no stake in the outcome is evidence, and evidence is what safety committees, insurers and regulators ask for.
Forge is that software. It builds a calibrated twin of the site, runs the robot's own code through it unmodified, and grades the result against the limits that matter. OEM engineers get a root cause; deployers get a readiness verdict; nobody gets a green tick on faith.
What we are not
- Not a robot OEM. We have no robot to sell and no release to defend.
- Not a system integrator. We don't earn from the deployment going ahead.
- Not a simulator vendor. We build on NVIDIA Isaac Sim rather than compete with it; our layer is the evaluation, not the engine.
What we are: the software the verdict comes from, with no stake in what it says.
Who we are
People who have done this before, for software.
Fireloop was started by a founder who has spent a career on one problem: how do you prove that software works before it reaches the people who depend on it?
The founder
Two companies built on testing and certification, both acquired
The first automated the certification of mobile apps for the major app stores: an independent check that an app was safe and ready before it reached the public, at a scale no manual process could match. The second built test-management and test-automation tools used by engineering teams to plan, run and track their testing on every release. Both were acquired by a Goldman Sachs private-equity portfolio.
Fireloop is the same discipline aimed at a harder target. The app store is now the warehouse floor. The app is a robot's driving policy. And the cost of a miss is no longer a bad review.
The team
Simulation engineers certified on the NVIDIA stack
Our engineering team is trained and certified on NVIDIA Isaac Sim, Isaac Lab and Omniverse, and builds Forge on that foundation. Fireloop is a member of NVIDIA Inception.
We build the way we ask our customers to: the AI drafts, a person reviews, and nothing ships without a graded result behind it.
How we work
Four commitments, each one visible in the product.
Values are easy to write. These are the ones Forge is built to keep, and where on this site you can see each one kept.
We never touch your software
Forge models the vehicle and its sensors. Your policy connects as a black box and makes every decision itself; weights and code stay on your network.
See it on the OEM pageA person approves every step
Forge's AI drafts the scene, the robot setup, the sensors and the scenarios. Your engineers review, correct and decide what counts as a pass.
See how Forge builds your worldEvidence, not a green tick
Every result carries the measurement that decided it, the margin to the limit, and a deterministic replay for your safety review.
See the verdictWe won't bend a safety test
When a detail is unclear, Forge's AI asks. When a change would alter what a test means, it says so. A pass you argued for is not a pass.
Why ForgeAt a glance
- Company
- Fireloop
- Founded
- 2025
- Headquarters
- San Jose, CA · engineering in India
- Product
- Fireloop Forge - simulation-based evaluation for physical AI
- First market
- Mobile robots that move goods: forklifts, pallet movers, tuggers, AMRs
- Built on
- NVIDIA Isaac Sim · Isaac Lab · Omniverse
- Standards referenced
- ISO 3691-4 · ANSI B56.5 · VDA 5050
- Programs
- NVIDIA Inception
Work with us
Three ways in.
Whichever side of a deployment you're on, the first conversation is the same: bring one scenario that worries you and we'll run it in Forge with you.
Building robots
Book a demo for OEMs
Bring your URDF and one site that gave you trouble.
For OEMsDeploying robots
Book a demo for deployers
Bring the aisle, dock or shift that worries you most.
For deployersPartners & press
Talk to us
Integrators, standards bodies, insurers and anyone writing about physical-AI safety.
build@fireloop.ai