Tangram Vision | MetriCal Production Line Calibration

Bring trust to Physical AI

Physical AI learns from real-world data, and that data is only as good as the sensors that capture it.

MetriCal is the calibration layer of the stack: it makes every downstream perception decision trustworthy by making the sensors agree with each other first.

From R&D to production

The first calibration is a test; the next ten thousand shouldn't be. MetriCal's steady-state workflow turns one good calibration into a factory process. Refine each new unit in a fraction of the time. Pipelines are TOML manifests: scriptable, versionable, CI-ready.

Kodiak

Calibration time cut to 13% of original, with one-button calibration running at contract manufacturers.

Saronic

MetriCal helped scale Saronic's production from one boat a week to one boat a day.

One pipeline, every modality

Intrinsics and extrinsics, space and time constraints; everything is solved jointly in a single optimization, not a chain of pairwise scripts that compound error.

Camera

Any number of cameras, calibrated together.

Local navigation

Mocap and tracking systems relative to the rest of the rig.

LiDAR

Multi-LiDAR rigs, no camera required in the loop.

IMU

Spatiotemporal calibration with preintegration error reporting.

Radar

Radar-inclusive rigs, shipping today.

Hand-eye

Robotic arms calibrated relative to their sensors.

Specialized coverage for interesting rigs: narrow field-of-view cameras with multiple targets, rolling shutter, thermal cameras, and multi-unit Intel RealSense setups.

Fastest cal in the west

Don't let your production rates be limited by your factory calibration. MetriCal delivers accuracy at the speeds demanded at scale, across all supported modalities.

I love Tangram. The technical whitepapers, blogs, and reference guides are incredible. I can’t believe how much they open source. I evangelize MetriCal everywhere.

Alex Carillo

Astrolab

Works with any lens you can bolt on

Eleven camera intrinsics models (and counting!), encompassing everything from narrow FOV to omnidirectional.

No Distortion OpenCV RadTan OpenCV Fisheye Fisheye Tangential (Tangram Model) OpenCV Rational Inverse Brown Conrady Inverse Kannala-Brandt EUCM Double Sphere Omnidirectional Power Law (Tangram Model)

Every price, published

From free trial to fleet-scale rate cards, everything is clear with MetriCal pricing.

[Trial Bundle

For new users

and the calibration curious](/content/pricing/index.html) [15 Credits

$249 each

15 Credits · $16.60 / Credit](/content/pricing/index.html) [R&D

$999/month

75 Credits / mo · $13.32 / Credit](/content/pricing/index.html) [Growth

$3,149/month

300 Credits / mo · $10.50 / Credit](/content/pricing/index.html)

Metrics you can defend

A calibration number without evidence is a liability. Every run produces a multi-section report: coverage, model fit, reprojection RMSE, camera-LiDAR misalignment, IMU preintegration error, per-component extrinsics deltas, and a go/no-go verdict. An HTML report for humans, MCAP for machines, and URDF conversion for your robot.

Blocking bad results

When MetriCal detects high-risk diagnostics, it will signal the operator and block results. You will never ship a bad calibration because nobody read the log.

Failed means free

Credits are only consumed by calibrations that pass quality thresholds. A failed calibration costs you nothing.