An autonomous mobile robot carrying a wrapped pallet down a warehouse aisle
Halden
Start free

Break it in simulation, not in the field

Halden runs every policy change against thousands of scenarios in the cloud before it reaches a robot, so you find the failure at the pull request, not on the floor.

Start freeRead the docs
Running on every pull request at
  • Toyota
  • Bosch
  • Hyundai
  • UPS
  • DoorDash
  • Volkswagen

You’re testing robot behaviourthe way software teams tested code in 2005.

Booking hardware to validate a change made sense when fleets were small and releases were rare. Neither is true anymore.

Before: an empty robotics test lab with traffic cones and taped floor markingsbefore
with halden

2,400

scenarios per change

A lab rig runs what fits in a night. Halden runs the full library, plus every incident you’ve replayed, on every pull request.

10 min

to a verdict

No lab booking, no robot to reset, no GPU queue. The sweep finishes in the time it takes to review the change.

the PR

where regressions surface

Failures arrive as a diff against your last known-good policy, with video, before anything ships.

Connect, pick, run, gate.Four steps to a policy you can trust.

  1. 1

    Connect your repo

    Point Halden at your repository and bring your robot model. Works with ROS 2 and the physics backends you already use; Halden runs on top of your simulator, it doesn’t replace it.

    read the setup guide
  2. 2

    Pick your scenarios

    Start with 2,400 pre-built scenarios across warehouse, aerial and manipulation tasks. Add your own by describing them, or replay a real incident straight from a rosbag.

    browse scenarios
  3. 3

    Run on every change

    Each pull request triggers a sweep executing in parallel in the cloud. A full 2,400-scenario run finishes in the time it takes to review the PR.

    watch a sweep
  4. 4

    Gate the deploy

    Results come back as a pass/fail diff against your last known-good policy. Below your threshold, the deploy is blocked and the failing scenario is waiting with video attached.

    set up a gate

From commit to verified policyin ten minutes.

A working setup end to end: connect a repo, run 2,400 scenarios, watch six fail, open the video of the one that matters. No slides.

halden — sweep view

One pipelinefrom commit to verified policy.

Everything between a policy change and a verified deploy runs in one place, so nothing gets lost between the lab, the logs and the pull request.

A floodlit warehouse behind a perimeter fence at dusk

Your models neverleave your control.

Robot models and field telemetry are among the most sensitive things a robotics company owns. We treat them that way, and we never train on them.

Data handling
  • SOC 2 Type II
  • Encrypted at rest and in transit
  • GDPR compliant
  • Deletable on request
Deployment
  • Self-hosted runners
  • VPC deployment
  • EU data residency
  • SSO and SAML
What we don’t do
  • Never trained on your models
  • No telemetry resale
  • Pen-tested annually
  • Audit log on every run
Legal signed off in a week. Our models run in our own VPC and never leave it.
Tomás Vargavp engineering, northbay amr

Run it onyour own robot model.

500 simulation-hours free every month. Connect a repo and see your first scenario result in ten minutes, no sales call.

Start freeWatch a demo
An engineer kneeling beside a quadruped robot, typing on a laptop

Runs on the stackyou already build with.

ROS 2 and the common physics backends, triggered from your existing CI. Reads rosbags directly, so a real incident becomes a scenario in one command.

  • GitHubci
  • GitLabci
  • Jenkinsci
  • ROS 2middleware
  • Foxglovevisualisation
$ halden replay ./bay7_2026-08-14.mcap --name bay7-near-missrosbag becomes a regression test

Numbers we’ll show youbefore you sign anything.

Measured across 1,100 policy changes with design partners. Methodology published in the docs.

Overhead view of warehouse racking with robots moving between the aisles
94%regressions caught before hardware
2,400scenarios in the library
10 minfrom signup to first run

What our customers say.

Priya Raghavan being interviewed
“We had two engineers maintaining a simulation rig that ran forty scenarios overnight. Halden runs twenty-four hundred on every pull request, and we gave those engineers back to the autonomy team.”
Priya Raghavanhead of autonomy, wayfarer robotics

Questions.Answers.

Everything teams ask before their first run. More detail lives in the docs.

  • No. Halden runs on top of the physics backend you already use and reads your existing robot description. Your scenarios and assets stay in the formats your team knows.

  • Most teams have a first sweep running the same day. Connect the repo, point Halden at your robot description, and the first 2,400-scenario run comes back in about ten minutes.

  • Accurate enough to catch regressions before they reach hardware: across 1,100 policy changes with design partners, Halden caught 94% of them. The methodology is published in the docs.

  • That’s the usual case. Halden works from your own robot description on your own physics backend, so a custom platform runs the same way a standard one does.

  • Yes. SOC 2 Type II, encrypted at rest and in transit, with self-hosted runners and VPC deployment if models can’t leave your network. We never train on your models or resell telemetry.

  • 500 simulation-hours are free every month, with no credit card. Beyond that, pricing scales with the simulation-hours you use. See the pricing page, or talk to us.

A truck crossing a highway overpass at dusk

Find it in simulation first.500 simulation-hours free every month.

No credit card, no sales call.

Start freeTalk to us