Add configurable simulation catalog, CAD import, and robot workflows

[moeuu]

What changed

• Add English-only local natural-language robot-task control with strict plan validation, confirmation gates, and audit output.
• Convert every tracked UI label, status message, prompt, example, test fixture, and documentation page to English.
• Remove the Japanese parser fallback, language variants, and Japanese font-atlas startup configuration.
• Add a repository regression test that rejects CJK text in any tracked text file.
• Place the LLM instruction section above explicit system configuration controls while keeping the two functions separate.
• Add a central catalog with independent environment, robot, and detector selectors, semantic CAD spawn anchors, and Linux SolidWorks-to-USD conversion.
• Add a permanent operator monitor: one explicit active robot, position/action/progress, one-click follow and onboard views, a 12 Hz building overview, routes, targets, detector measurement positions, contamination geometry, and through-wall robot beacons.
• Cancel automatic follow on manual camera edits and make exactly one smooth work-view transition after physical contact is confirmed and decontamination starts.
• Connect Ridgeback+Franka and Nova Carter controller progress to the UI, including approaching, contact-confirmed, treatment coverage, and completion phases.
• Render treated coverage from the exact irregular activity-bearing faces used by contact validation; no rectangular or hidden coverage proxy is introduced.
• Keep the main viewport at the existing 60 FPS cap and reuse that one viewport for the selected onboard camera instead of rendering every robot camera.
• Make the Operations panel vertically scrollable so monitoring, LLM, configuration, measurement, and workflow controls remain reachable.

Why

RadCounterSim is intended for operators who should not need commands, USD paths, spawn coordinates, or camera controls. Imported CAD and catalog components must remain independently selectable, while the operator still needs to understand which robot is working, where it is in the building, what it will do next, and how much of the physical task is complete. The repository and product interface are now consistently English-only.

Impact

Operators can select any catalog robot with one click, follow it from a stable diagonal rear view, switch to its onboard view, or return to a building overview. The viewport and minimap show the current route, target, heading, distance, detector measurement location, and the real irregular contaminated geometry. During decontamination the status advances from approach to contact confirmation to live treatment progress, and the camera moves once to a view containing both robot and work surface.

The articulated CAD decontamination gate requires the matching central-catalog robot set, so an unselected task robot can no longer appear under a PackBot/Elios roster. Natural-language plans and runtime metadata now advertise and accept English only.

Validation

• uv run ruff check .
• uv run pytest -q — 213 passed, 3 skipped because the optional Embree native module is not built
• git diff --check
• Repository CJK audit — zero matches in tracked text files
• Visible Isaac Fukushima camera gate — follow, onboard, work, manual-cancel, and overview modes passed; overlay error null; 12 Hz; one rendered viewport
• Visible Isaac Fukushima articulated decontamination gate — 0.0 m spawn drop, 68,246 accepted contacts, 93.36% exact irregular-face coverage, 31.68% activity removed, 12 raster points, 95 contamination map points, active robot countermeasure / Ridgeback + Franka, work camera, no extra validation facility