Physical AI acceptance workbench

Same map. Day and night.
Prove the invariant.

Compare measured localization runs against one explicit deployment contract. See every failed gate before a robot or pair of glasses enters the building.

Evaluate measurements See the evidence model

This tool evaluates measurements you provide. It does not perform image localization or connect to Multiset.

One acceptance contract across the clock.

Enter measured output from the same mapped space. Every value is compared directly with the gates you set.

Reference condition

Noon run

Coverage% Inlier ratio% Position errormeters Confidence0–1

Changed illumination

Midnight run

Coverage% Inlier ratio% Position errormeters Confidence0–1

Deployment gates

Min coverage% Min inliers% Max errorm Min confidence0–1

Noon

Midnight

Changes save automatically in this browser.

Lighting changes. The acceptance logic doesn't.

Each decision remains traceable to a measured input, a named gate, and a deterministic comparison.

Reference the same physical map.

Keep the space fixed. Treat illumination as the changed condition, so the benchmark tests invariance instead of moving the target.

Measure the difficult condition.

Record midnight coverage, geometric inliers, position error, and confidence from the localization system you actually use.

Carry the evidence forward.

The export preserves all inputs, gates, individual checks, and the final decision for review or CI ingestion.