> ## Documentation Index
> Fetch the complete documentation index at: https://kensou.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

# Crossing · 跨 embodiment 合流 ★ 本書 USP

The five dimensions every embodiment touches but no single survey compares. **The reason this handbook exists.**

| Subdirectory                | What it answers                                                                                    | First wedge                                                                                                  |
| --------------------------- | -------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------ |
| `scale-comparison/`         | 1 cm → 1000 km: how the same problem deforms across scale                                          | `depth-foundation-across-scales.md`                                                                          |
| `sensor-stack-matrix/`      | Why each embodiment picks the sensors it does (SWaP-C maths)                                       | **`sensor_budget_matrix_v1.md` (W2)**                                                                        |
| `slam-vio-migration/`       | Why VINS-Mono works on EuRoC, breaks underwater, overkills the desktop                             | **`vggt_vs_drone_vio.md` (W1) — flagship piece** · `orb_slam3_vs_vins_fusion_code_comparison.md` (代碼層工程哲學對比) |
| `representation-migration/` | 3DGS as simulator: manipulation vs drone vs AD · 跨表征 3D registration                               | `3dgs-as-simulator-comparison.md` · `dense_vs_graph_registration.md` (★ NEW — 5 法 registration 对照)           |
| `failure-modes-atlas/`      | Transparent / reflective / deformable / weather: same failure, different mitigation per embodiment | `transparent-reflective-deformable.md`                                                                       |

Each piece targets 10k–20k Chinese chars depth, citing 8–15 papers from across embodiment lanes.