Benchmark Reference#
This section documents every benchmark in the bundled FET suite: what each test checks, what a passing result guarantees, how the test works, the failure modes, and how to fix a failing asset. Tests are organized by feature: each family page lists the tests that validate that feature.
What the Framework Does#
Benchmarking runs as a pipeline: the planner selects the tests an asset is eligible for, the runner executes them in the engine and captures frames and logs, the stamper records the outcome on the asset, and the reporter aggregates results into an HTML and JSON report. Refer to Pipeline and Stages for the full pipeline, and to Running Tests to set up and run the tool.
How to Read a Test Result#
Each feature in a report carries one state. A feature is PASS when at least one of its tests passed or was skipped and none failed; a feature whose applicable tests are all skipped, such as a world-anchored asset, is reported as PASS. A feature is FAIL when any of its tests failed. A feature is NEUTRAL when it is validated but has no benchmark to run. A feature is VALIDATION_FAILED when it did not pass static validation, so no benchmark was run for it. Refer to Reading Reports to interpret a full report.
Test Families#
Family |
Validates |
|---|---|
Visual presence, surface normals, back-face culling, lighting response, and pivot placement. |
|
Rigid-body ground drop, slope drop, and resting stability. |
|
Multibody articulation and joint movement. |
|
Grasp, lift, shake, and release. |
|
Driven joint range, velocity, effort, mimic behavior, coordination, and inverse kinematics. |