How it works
From site scan to signed verdict
Onboarding takes weeks, not quarters. After that, every policy update is validated automatically in the background of your release process.
nav-policy v2.9.0
- Site
- DC-04 twin
- Scenarios run
- 12,480
- Collisions
- 0
- Min. human clearance
- 0.62 m
- p99 stop time
- 212 ms
- Est. incident rate (95% CI)
- < 1.3 × 10⁻⁶ / h
The process
Six steps from first call to steady state
Week 1–2
Capture the site
Share CAD, LiDAR or photogrammetry scans and a sample of fleet logs. Twinora builds the twin.
Week 2–3
Calibrate the twin
We replay real logged runs in the twin and tune it until outcomes match.
Week 3
Pick the suites
Choose scenario suites from Scenova and add incidents from your own history.
Week 4
Connect the pipeline
Add the CLI step or webhook to CI, so every candidate build triggers a sweep.
Ongoing
Gate releases
Veridex issues a verdict for each build. Passes roll out, fails come back with replays.
Ongoing
Close the loop
Sentryx streams real outcomes so the twin and the library keep improving.
Who does what
Clear ownership from day one
| Task | Your team | Simprova |
|---|---|---|
| Site scans and floor plans | Provide existing files or site access | Process, clean and build the twin |
| Robot models (URDF / USD) | Provide or approve | Import, validate and tune physics |
| Fleet logs | Export a sample (rosbag2 or MCAP) | Use for calibration and scenario mining |
| Pass criteria | Set with your safety lead | Encode as a versioned suite |
| CI integration | Add one pipeline step | Provide templates and support |
| Ongoing sweeps | Review fails | Run, score and store evidence |
Pass criteria
You define what safe means
Every suite is a versioned file your safety lead signs off. Change a threshold and the change is tracked like code.
- Minimum human clearance and stop distances
- Collision and contact force limits
- Task success and throughput floors
- Allowed failure rate with a confidence level
suite: release-gate version: 7 twin: site-warehouse-b scenarios: - scenova/human-intrusion/* - scenova/sensor-dropout/lidar-* - site/incidents-2026 variations_per_scenario: 24 criteria: collisions: 0 min_human_clearance_m: 0.5 p99_stop_ms: 250 task_success_rate: ">= 0.97" incident_rate_per_h: "< 1e-5 @ 95%"
What you get back
A verdict you can act on
Pass
Signed certificate
Rollout is unblocked. The certificate is attached to the release for audit.
Fail
Failing cases with replays
Each failing run opens in 3D with the seed, sensor view and the policy's actions.
Review
Borderline results
When a result sits near a threshold, Veridex flags it for a human decision instead of guessing.
Questions
About onboarding
What if we don't have LiDAR scans?
CAD plus a walk-through video is enough to start. We can arrange a scan if fidelity needs it.
How do you know the twin matches reality?
We replay logged real runs in the twin and compare outcomes. The twin's fidelity score is shown on every certificate.
How long does a sweep take?
Most release-gate sweeps finish in one to four hours, depending on suite size and GPU allocation.
Can we run ad-hoc sweeps outside CI?
Yes, from the dashboard or CLI, for experiments and research builds.
Start with one site
The pilot program covers one site, one robot type and one release cycle.