Turing Sim: Making the USD Schemas Inspector Readable

The running editor shows separate USD schema cards, UR10 articulation settings and the source asset window.

中文版在文末，点开即可阅读。

Each USD schema on the selected cube (Collision, Mesh Collision, Rigid Body, Mass) now sits in its own card with its USD schema name.

The selected UR10 shows Translate, Rotate, Scale and Articulation Root settings in the running editor.

The source asset window shows a hierarchy, preview and Properties/Raw USD tabs.

Situation

Turing Sim is an experimental OpenUSD editor and robotics simulation workbench. Its Schemas tab lists every schema on the selected prim, including physics APIs, and every edit goes through the document's undo history. On September 28 the tab worked, but it was hard to read. A viewport tool section sat between the name and the Transform, schemas without properties printed "No additional properties", and the Rigid Body and Mass fields ran together with no visible boundary. Transform showed only the transform parts already stored in the file, so the UR10 showed Position but no Rotation or Scale.

Task

The goal was to make the tab read the way USD is structured, following the conventions of Isaac Sim's Property panel. Every property should clearly belong to one schema, Transform should own translate, rotation and scale, and the Articulation Root should expose settings such as self collision. The difficulty was doing this without hardcoding robot or schema names, which the project rules forbid, and without breaking undo.

A second problem appeared during testing. After moving the UR10 with the gizmo, the Transform section switched to "custom transform stack" and the position fields went stale.

Action

1. One card per schema. Each schema is now a bordered card. Its header shows a readable name on the left, the USD schema identifier (for example PhysicsRigidBodyAPI) on the right, and a remove button for applied APIs. Removing a schema also clears its values in the current edit layer, as one undo step. Cards that are mostly arrays, such as mesh topology, start collapsed.
2. Isaac Sim-style rows. Rows are two columns, label and value, and stack when the panel is narrow. Labels come from the schema's own displayName metadata, such as "Center of Mass". A dot at the end of each row shows whether the value is authored: hollow for the schema default, blue when set in the current edit layer (click to revert), grey when set in another layer. Booleans are checkboxes, and physics fields whose USD default is infinity read as "Auto".
3. Transform owns position, rotation and scale. Translate, Rotate (or Orient) and Scale always appear, named after their USD transform ops. Rotation switches between Euler degrees and a quaternion. Either input is converted to whichever form the prim already stores, so Isaac-style orient quaternions stay quaternions.
4. The gizmo edits the real transform. The cause of the stale fields was that drags wrote a separate matrix op instead of changing the prim's own translate and rotate values. Drags now update those values directly. The matrix op remains only as a fallback for transforms that position/rotation/scale cannot represent. Older matrix ops are merged back on the next edit.
5. Articulation Root settings. The card counts the rigid bodies and joints found under the root and offers the PhysX articulation settings: enabled, self collisions, solver iterations, and sleep and stabilization thresholds. Editing one writes the value and applies the PhysX articulation schema in the same undo step.
6. Source asset editor. The separate editor for referenced assets now uses the same glass cards and segmented tabs as the main window. A user-provided screenshot of the native editor shows the revised layout; editing behavior in that window has not been checked in a headed session.

Regression tests cover each new edit path with undo and redo, plus a check of all six Euler axis orders against USD's own rotation ops. The full offscreen suite gave 525 passed, 3 failed and 6 skipped. The three failures were older tests still looking up the field by its previous name, "Position X". After updating them, the affected test files pass (61 tests). The whole suite has not been re-run since that fix. The three screenshots at the top show the running app after the changes: separate schema cards, the UR10 Transform and Articulation Root panels, and the source asset editor. They show visible layout and values, not an end-to-end edit or save. The other checks were headless, and today's changes are not yet committed.

Result

In the running editor, a cube with collision, rigid body and mass now shows four clearly separated cards, and each field sits under the schema that owns it. Gizmo moves now show up in the Transform fields, and the UR10's articulation settings can be edited in place.

Next steps are to commit and push these changes, re-run the full suite, test the gizmo against real robot assets in the native viewport, and check the Translate fields at narrow panel widths.