[TATP-233]

Summary

• Document and regression-test named <actuatorfrc> sensor reads through the existing packed Tensor IO path, including actuator ordering and selected refreshes.
• Add physical-substep delay_min_lag / delay_max_lag to joint-position actions, including motion-tracking actions, Manager RNG, selected reset, and nonfinite validation.
• Add stable runtime terrain origins/levels/types and reset-scoped terrain_levels_by_progress, consuming UniSim's public terrain spawn metadata without rebuilding geometry.
• Update English and Chinese migration and terrain documentation. Default task behavior is unchanged; training changes only for owners that enable command delay or terrain curriculum.

Linked Work

• Issue: Closes 请求补充执行器力矩读回接口（参考 mjlab actuator force） #2562, Closes 请求恢复 DC 电机执行器配置能力（反射惯量/armature/指令延迟，参考 mjlab DcMotorActuatorCfg） #2563, Closes 请求补充地形运行时选级接口（terrain_origins/terrain_levels，参考 mjlab TerrainEntityCfg） #2564
• Parent roadmap (when applicable): none
• Roadmap declared base (when applicable): N/A
• Milestone: N/A
• Base branch: main
• Governing contracts: ADR-0007, ADR-0012, ADR-0015

Validation

• make test-all passed on the final local head before this PR was created or updated
• Additional task-specific validation listed below

Commands actually run:

make test-all
uv run pytest tests/scripts/test_check_docs.py -q
cd docs/sphinx && UNILAB_DOCS_SKIP_AUTODOC=1 uv run --no-project --with-requirements requirements.txt sphinx-build -b html -n source build/html

Results on e071c492d1b001b2984288e5f7cad6b7ee5d679e:

• make test-all: 2937 passed, 40 skipped, 375 deselected, 1 xfailed; benchmark smoke 35/35 module imports and 36/36 script imports.
• Documentation contract checks: 20 passed.
• Sphinx: passed.

Remote CI route:

• Base main: all current-head CI and Docs checks passed on e071c492d1b001b2984288e5f7cad6b7ee5d679e, including coverage, benchmark smoke, all test suites, type checks, Ruff, and Sphinx.

Impact

• Backend impact: both through the existing public terrain metadata contract; MuJoCo numerical integration coverage; no backend adapter changes
• Platform impact: Linux validation; no platform-specific code
• Training effect expected: yes, only for task owners enabling command delay or terrain curriculum; zero-delay and non-curriculum defaults retain existing behavior

Artifacts

• W&B: N/A
• benchmark result: repository benchmark smoke only; no performance benchmark claimed
• video / screenshot: N/A
• ONNX / checkpoint: N/A

Checklist

• Added or updated tests where needed
• Updated docs if behavior or workflow changed
• Linked the driving issue
• Noted any follow-up work explicitly

Follow-up boundaries: no Motrix-specific runtime curriculum run was performed, and full DC torque-speed saturation is outside the requested armature/gain/delay scope.

Companion UniSim regression PR: unilabsim/unisim#494.
Current-head remote runs: CI, Docs.