Navigation quality and simulation fidelity upgrade foundation

[AoiOTA]

Summary

• deliver the staged navigation-quality and simulation-fidelity foundation: Warehouse V2 registration, motion baselines, physics-step ROS publication, camera controls, lifecycle/reset fixes, managed-process shutdown, versioned robot kinematics, contact profiles, topology isolation, and strict evidence aggregation
• bind versioned Reset A/B identity and the wheel-to-ground contact probe into runtime provenance v6, motion schema 3, the ordered 47-column Manifest v2, analysis 5, physical 3 / policy skid_steer_plan_8_7_v3, and summary 6
• persist root-pose restore through the USD backend plus PhysX flush and articulation synchronization; alternate A/B then B/A cold-process order to counterbalance startup order
• disable default viewport updates in headless mode while retaining RTX sensor render products, preventing sequential-run descriptor exhaustion without suppressing Warehouse LiDAR
• provide a beginner-oriented Chinese user manual and a complete 320/320 repository file index; synchronize README, interfaces, plans, and verification evidence

Verification

• current HEAD 9c597e8: pytest -q completed with 1270 passed and 34 skipped
• clean 65ae923: ./scripts/test.sh --with-isaac exit 0
  • root: 1268 passed / 1 skipped / 35 deselected
  • ROS workspace: 1035 tests / 0 errors / 0 failures / 1 skipped
  • Isaac/USD marker: 32 passed / 1 skipped / 283 deselected
• clean 65ae923 Reset A/B smoke: 2/2 runs completed with the current 47-column and 3/5/3/6 report chain
• clean 65ae923 formal Reset A/B: 20/20 runs, 20 included / 0 excluded, two groups with 10 repeats each, and 120 validated segments
  • A pose_restore_v1: 5/10 repeats passed physical acceptance
  • B separate_recontact_0p20m_1step_v1: 4/10 repeats passed physical acceptance
  • the only failed leaf was rotation_center_drift_asymmetry_ratio; all other physical categories passed 10/10 in both groups
  • manifest da7faad19cbcb247287ae241626bd392562e6737dca6d0fa91f82eac236121e5
  • analysis 9e595e51dd625718d45bf1a4d8611248eb8ee13daf720ab0d8764ed814b1deba
  • summary 1b7778236594fab4867f5d04956dcb1f7955144f4e868f3835c63911930f72d7
• Warehouse headless RTX smoke after the viewport fix: 77 /lidar/points_raw samples in 8 seconds, 9.6028 Hz, 38,264,444 bytes; camera topics remained disabled
• documentation audit: 84 local links valid; repository index tracked/indexed/unique 320/320/320 with no missing, stale, or duplicate paths; git diff --check passed

Current acceptance boundary

This PR remains draft and does not claim completion of the full plan. The formal Reset A/B evidence chain succeeded, but both physical groups failed the rotation-center-drift asymmetry gate. Strategy B performed worse than A and is not promoted; pose_restore_v1 remains the default.

Required later gates still include the complete topology matrix, multi-speed and 60/120 Hz validation, Realistic navigation replay, Warehouse V2 calibration and route validation, camera visual/RTF/GPU revalidation, N01-N21, real RViz N19 shutdown cycling, and final static/dynamic statistical runs of at least 100 trials each. N20 and N21 currently each have 0 formally countable samples.

Promote and merge only after the remaining required gates are complete, or after the project scope is explicitly split into separately accepted phases.