Real logs in.
Adaptable sims out.

Feed Brio your logs. Its world models reconstruct the run as a simulation you can replay, adapt, and stress-test, covering the field scenarios you could never record.

Sensor logs · World models · VLAs · Simulation · Autonomy

Patrick Walsh

"At previous startups and an aerospace R&D company, I watched teams pour effort into hand-building sim worlds, and still get blindsided by scenarios they never thought to author. We had the field logs the whole time. I built Brio to turn them into sims you can actually test against."

Patrick Walsh·Co-Founder & CTO
Sensor logslogs
Pick the runs to rebuild as simulations.
BRIO/ˈbrē·ō/
noun
vigorandvivacityofstyleorperformance;spiritedenergy.
· Italian · 19th c. · from brio "liveliness"
What is Brio

Your logs become simulations you can stress-test.

Brio is built for robotics teams. Point it at your sensor logs from a real or simulated run, and its world models reconstruct that run as a simulation you can execute: the same sensor streams, the same transforms, the same dynamics, now runnable.

From there, change the world. Inject the edge cases you never recorded, like a wet floor, a dropped frame, or a pedestrian crossing late, and re-run. You test against scenarios that never made it into a log, without staging them on real hardware.

Brio also debugs the sims you already have. Point it at a scenario that crashes or drifts from the logged run, and it traces why: a mismatched sensor model, a divergence from the field data, instead of leaving you to guess.

The problem3 known issues
01

Building sim worlds by hand is slow

Every scenario is modeled by hand: geometry, dynamics, sensor placement. The effort scales with the long tail, and the long tail never ends.

Another scenario to author from scratch before you can test against it.

02

The scenarios you need are unrepeatable

The edge cases that break robots happen once, in the field, and never the same way twice. You can't re-run reality, and you can't author what you didn't anticipate.

The robot did something strange near the loading dock. Good luck reproducing that in a hand-built sim.

03

Sims drift from reality, silently

A scenario passes in sim, then the robot behaves differently on hardware. Without comparing against the logged run, you don't find the divergence until the field does.

It worked in sim. It didn't in the field. Nothing told you why.

Industrial robotic arm in dramatic lighting
CAPABILITY · 01
logs://sim
RVO · 01 / 05·· FRAME 2048
── Capabilities

Built for the
real stack.

Early access

Your robot.
In the field.

The field is where the scenarios that matter live. Brio turns the logs from those runs into sims you can replay and stress-test. Early access for teams putting autonomous systems through their paces.

Frequently asked06 questions

Questions,
answered.

The things robotics teams ask us most often: what Brio is, how it differs from Claude Code, and what makes it ours.

Ship your robots
tested against the real world.

Join engineers building the next generation of autonomous systems. Early access is limited.

Brio Workspace
brio · ~/drone_nav_ws
live
should I get brio?
Brio

I read the logs from your last field runs.

I rebuilt each as a runnable scenario, then varied the conditions you never recorded: wet floor, dropped LIDAR frames, a pedestrian stepping in late.

Your stack held on the originals. Some of the generated scenarios surfaced a planner edge case, caught in sim, before it reached hardware.

Recommendation: Definitely.
Backed by
LAUNCHNVIDIA Inception ProgramCreateX Launch