TrialONE — break your robot in simulation

What we do

We break your robot on purpose, in sim.

TrialONE is a local desktop app. It stress-tests your robot in simulation and, when it crashes, the built-in AI pinpoints exactly where it failed, how, and what to change.

From injected faults to patterns

illustrative

How we do it

One scenario, a handful of functions.

Core

Fault injectionfrom the outside

Drop a topic, spike latency, kill a node, spoof GPS — on a schedule. Your robot's code is never touched.

drop_topicnode_killlatency

Core

AI crash analysis

When it fails, the AI pinpoints where it broke, the root cause, and the fix to try.

wherehowfix

Core

Independent oracle

PASS/FAIL judged from simulator ground truth — never the robot's own estimate.

ground-truth

Local

Runs 100% local

A desktop app on your machine. No cloud — your robots and data never leave.

desktopprivate

Open

Scenario format

A small declarative YAML describes what to run and how to break it.

scenario.yaml

Core

Crash recording

A machine-readable run_record.json plus an auto before/after crash.mp4.

run_recordcrash.mp4

What it looks like

A scenario, the run, the verdict.

A small YAML declares what to run and how to break it. The CLI runs it. You get a machine-readable verdict — PASS/FAIL from independent ground truth.

name: lidar-dropout-amr
sim:    { backend: gazebo, world: worlds/walled_room.sdf }
robot:  { model: turtlebot3, stack: nav2 }
goal:   { type: reach_pose, x: 4.0, y: 0.0, tolerance: 0.3 }
faults:
  - kind: drop_topic
    topic: /scan        # blind the lidar
    at: 8.0s            # 8 s into the run

How it helps you

Catch the crash before the field.

Failures surface on a re-run in sim — and the AI hands you the cause and the fix — instead of hunting them down slowly, expensively, sometimes dangerously in the field.

Robotics teams

Crash & safety regression testing before field deployment.

ML / perception researchers

Stress perception and navigation against sensor degradation.

CI engineers

Block the merge if the robot crashes in sim.

Who we are

Two builders, honestabout what's real.

Pre-alpha. Two founders (TUM / LMU). The design is ahead of the code — and we say which is which.

Harwin Singh Malhi

Co-founder · AI & systems · LMU

AI systems & agentic LLM workflows. Builds the fault-injection engine and the independent oracle.

LinkedIn →

Philip Degenhardt

Co-founder · Aerospace · TUM

Aerospace & CAD at TUM. Builds the scenario format and robot/sim integration.

LinkedIn →

Want to try it?

Leave your email for early access — or book a call, walk us through your use case, and we'll set you up with the app.

Pre-alpha. Two founders (TUM / LMU). The design is ahead of the code — and we say which is which.