File size: 13,200 Bytes
79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 89f2844 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 96ac0b2 79e17fd 0cc25df 96ac0b2 89f2844 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd 4b6da0e 0cc25df 4b6da0e 0cc25df 4b6da0e 7c7b56f 4b6da0e 0cc25df 0654d0d 4b6da0e 32773e7 96ac0b2 79e17fd 4b6da0e 79e17fd 4b6da0e 79e17fd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 | # Placement Assets 资产库
这是一个面向机器人仿真与 placement 任务的统一资产库,当前托管在 Hugging Face Dataset:
```text
https://huggingface.co/datasets/yuchi233/Placement_Assets
```
本仓库不是 RoboCasa 的镜像,也不是基于 RoboCasa 单独建立的资产库。RoboCasa、DISCOVERSE、TinyForge、Hunter 和第三方资产站都被视为不同的资产来源,统一纳入同一套存储、命名、索引和溯源规范。
## 当前已入库资产
- RoboCasa 洗碗机 fixture:
- 路径:`assets/robocasa/raw/fixtures/dishwashers/`
- 数量:25 个 MJCF 资产
- 每个资产保留 `model.xml`、`visuals/` 和本地 `metadata.yaml`
- RoboCasa 洗碗机相关辅助文件:
- registry:`assets/robocasa/raw/fixtures/fixture_registry/dishwasher.yaml`
- fixture 实现参考:`docs/source_notes/robocasa/dishwasher_fixture.py`
- DISCOVERSE models 集合:
- 路径:`assets/discoverse/raw/collections/models/discoverse_models/`
- 内容:MJCF、URDF、mesh、PNG/reference image 等
- 处理方式:作为 collection-level asset 入库,内部保持 DISCOVERSE upstream 的 `models/` 目录结构,避免破坏 MJCF include 和 mesh 相对路径
## 统一目录结构
原始资产使用 `raw/`,保留来源结构和相对引用:
```text
assets/<source>/raw/<asset_type>/<category>/<asset_id>/
```
示例:
```text
assets/robocasa/raw/fixtures/dishwashers/Dishwasher031/
```
派生资产使用 `derived/`,用于保存凸分解、collision mesh、清洗结果、格式转换等生成物:
```text
assets/<source>/derived/<asset_type>/<category>/<asset_id>/
```
如果要创建凸分解脚手架,可以先运行:
```bash
python scripts/scaffold_convex_decomposition.py assets/robocasa/raw/fixtures/dishwashers/Dishwasher031
```
大规模 upstream 资产树如果包含大量内部相对引用,也使用 `raw/` 结构,但可作为集合资产保存:
```text
assets/<source>/raw/<asset_type>/<category>/<asset_id>/
```
示例:
```text
assets/discoverse/raw/collections/models/discoverse_models/
├── metadata.yaml
├── mjcf/
├── meshes/
└── urdf/
```
## 命名规则
目录字段:
```text
assets/<source>/raw/<asset_type>/<category>/<asset_id>/
```
- `source`:资产来源,使用小写 snake_case,例如 `robocasa`、`discoverse`、`tinyforge`、`hunter`、`third_party`
- `asset_type`:资产类型,例如 `fixtures`、`objects`、`scenes`、`robots`、`collections`
- `category`:类别,例如 `dishwashers`、`models`、`cups`、`cabinets`
- `asset_id`:优先保留 upstream 稳定 ID,例如 `Dishwasher031`
全局 asset_id:
```text
<source>.<asset_type>.<category>.<upstream_id>
```
示例:
```text
robocasa.fixtures.dishwashers.Dishwasher031
discoverse.collections.models.discoverse_models
```
## 仓库结构
```text
.
├── AGENTS.md # 给 agent / 后续维护者的仓库使用说明
├── README.md # 英文 README
├── README_zh.md # 中文 README
├── assets/ # 实际资产文件
│ ├── discoverse/
│ ├── hunter/
│ ├── robocasa/
│ ├── third_party/
│ └── tinyforge/
├── templates/ # 元数据模板
├── docs/ # 规范、来源说明、调研记录
│ ├── asset_readiness_classification.md
│ ├── convex_decomposition_spec.md
│ ├── external_source_catalog.md
│ ├── naming_spec.md
│ ├── placement_asset_notes.md
│ ├── primitive_collision_proxy_spec.md
│ ├── structure_mapping.md
│ ├── source_notes/
│ └── storage_spec.md
├── manifest/
│ ├── assets.jsonl # 全局资产索引,一行一个资产
│ ├── licenses.yaml # license 字典
│ └── sources.yaml # 来源与候选第三方来源
└── scripts/
├── run_convex_decomposition.py
├── run_primitive_collision_proxy.py
├── scaffold_convex_decomposition.py
├── download_from_hf.sh
├── upload_to_hf.sh
└── validate_asset.py
```
## 索引与 metadata
全局索引文件:
```text
manifest/assets.jsonl
```
每个资产目录也必须包含:
```text
metadata.yaml
```
必需字段包括:
- `asset_id`
- `source`
- `source_asset_id`
- `asset_type`
- `category`
- `format`
- `entry_file`
- `license`
- `origin_url`
- `path`
- `tags`
推荐字段包括:
- `readiness_level`
- `source_commit`
- `retrieval_date`
- `conversion_tool`
- `conversion_notes`
- `raw_sha256`
- `validation_status`
## 原始结构与当前结构映射
如果导入时修改了 upstream 的目录结构,必须记录原始结构和当前结构的对应关系。
统一记录位置:
```text
docs/structure_mapping.md
docs/source_notes/<source>.md
```
当前例子:
```text
DISCOVERSE upstream:
/home/yuchi/projects/DISCOVERSE/models/
当前仓库:
assets/discoverse/raw/collections/models/discoverse_models/
```
RoboCasa 洗碗机则是一对一映射:
```text
robocasa/models/assets/fixtures/dishwashers/Dishwasher031/
→ assets/robocasa/raw/fixtures/dishwashers/Dishwasher031/
```
## 资产就绪度 L0–L5
资产存进仓库不代表可以直接用于产数。就绪度单独记录:
- L0:已通过目标任务、机器人 embodiment、policy、verifier 和 episode replay 验收,可稳定产数
- L1:在原仿真器中是完整可运行资产,但尚未适配目标产数链路
- L2:已有 articulated 结构,但物理、碰撞、任务语义等信息不完整
- L3:可动部件已分离,但缺少可靠 joint
- L4:完整静态整件
- L5:原始、缺件、未装配或需要显著修复的静态 mesh
完整标准见:
```text
docs/asset_readiness_classification.md
```
当前状态:
- RoboCasa dishwasher:`L1_candidate`
- DISCOVERSE collection:`L1_L2_candidate_by_subasset`
## 凸分解结果存放方式
凸分解结果属于派生资产,不能覆盖原始 mesh,也不能直接混进 upstream 的 `raw/` 目录。
统一存放路径:
```text
assets/<source>/derived/convex_decompositions/<category>/<asset_id>/
```
示例:
```text
assets/robocasa/derived/convex_decompositions/dishwashers/Dishwasher031_vhacd_v1/
├── metadata.yaml
├── source_refs.yaml
├── meshes/
│ ├── door/
│ │ ├── part_000.obj
│ │ └── part_001.obj
│ └── rack0/
├── mjcf/
│ ├── convex_assets_include.xml
│ ├── convex_geoms_include.xml
│ └── convex_collision_include.xml
└── logs/
└── decomposition.json
```
其中:
- `metadata.yaml`:记录派生资产 ID、license、来源、算法、参数文件、验证状态
- `source_refs.yaml`:记录对应的原始 asset、原始 XML、原始 mesh 路径
- `meshes/`:保存凸分解后的 convex pieces
- `mjcf/convex_assets_include.xml`:在 MJCF 顶层 include,用来注册 convex part mesh
- `mjcf/convex_geoms_include.xml`:在目标 body 内 include,用来添加 collision geoms
- `mjcf/convex_collision_include.xml`:带 wrapper body 的独立预览/便利 include
- `logs/decomposition.json`:记录 VHACD / CoACD 等工具版本、参数、输入输出 hash
- `logs/parts_summary.csv`:记录每个 part 的 bbox 尺寸、中心、体积、面积、顶点数和面数,方便筛查过小/过大的 collision part
实际运行 CoACD 凸分解:
```bash
pip install trimesh coacd
python scripts/run_convex_decomposition.py \
assets/robocasa/raw/fixtures/dishwashers/Dishwasher031 \
--mesh visuals/door.obj \
--mesh visuals/rack0.obj \
--variant coacd_v1 \
--register-manifest
```
MuJoCo 接入方式:
```xml
<!-- MJCF 顶层 -->
<include file="path/to/mjcf/convex_assets_include.xml"/>
<!-- 目标 body 内部 -->
<include file="path/to/mjcf/convex_geoms_include.xml"/>
```
不要直接修改 `raw/` 里的 `model.xml`。如果要替换原始 collision,应该在下游任务工程里选择性 include 派生 collision 文件。
凸分解后的 part 尺寸不能直接通过“目标长度/目标体积”指定,CoACD 主要通过粒度参数间接控制:
```text
--threshold 越小通常越细,part 更多/更小;越大越粗,part 更少/更大
--max-convex-hull 限制最多生成多少个 convex hull;越小越粗
--no-merge 禁止合并 hull,通常会保留更多/更小的 part
--mcts-max-depth 允许更深层递归切分,可能得到更细的结果
--resolution 提高几何采样/细节分辨率
```
粗 collision 示例:
```bash
python scripts/run_convex_decomposition.py <raw_asset_dir> \
--mesh <mesh.obj> \
--threshold 0.08 \
--max-convex-hull 8
```
细 collision 示例:
```bash
python scripts/run_convex_decomposition.py <raw_asset_dir> \
--mesh <mesh.obj> \
--threshold 0.02 \
--max-convex-hull 32 \
--no-merge \
--mcts-max-depth 4
```
完整规范见:
```text
docs/convex_decomposition_spec.md
```
该文档包含 `scripts/run_convex_decomposition.py` 的完整使用方法、所有 CLI 参数说明、粗/细分解参数 preset,以及 rack / wire-like 资产的注意事项。
## Primitive collision proxy 结果存放方式
对于洗碗机 rack、钢丝架、栅格、插槽隔板这类 thin rod / wire-like 资产,如果 CoACD / VHACD 会把功能空隙填死,优先使用 primitive collision proxy。
统一存放路径:
```text
assets/<source>/derived/primitive_collision_proxies/<category>/<asset_id>/
```
示例:
```text
assets/robocasa/derived/primitive_collision_proxies/dishwashers/Dishwasher031_rack1_primitive_boxes_v1/
├── metadata.yaml
├── source_refs.yaml
├── README.md
├── REPORT.md
├── proxy/
│ └── primitive_collision_proxy.json
├── mjcf/
│ ├── primitive_geoms_include.xml
│ └── primitive_collision_include.xml
├── curobo/
│ └── primitive_world.yml
└── logs/
├── cuboids_summary.csv
└── proxy_generation.json
```
实际运行示例:
```bash
python scripts/run_primitive_collision_proxy.py \
assets/robocasa/raw/fixtures/dishwashers/Dishwasher031 \
--mesh visuals/rack1.obj \
--variant rack1_primitive_boxes_v1 \
--segment-length 0.03 \
--padding 0.001 \
--min-half-extent 0.0015
```
其中 `proxy/primitive_collision_proxy.json` 是 source of truth,MuJoCo 和 cuRobo 输出都从同一份 JSON 派生。
完整规范和参数说明见:
```text
docs/primitive_collision_proxy_spec.md
```
## 第三方资产规则
第三方来源可以先只记录链接,不下载、不入 manifest。
只有当某个项目、实验、任务或生成场景实际使用了第三方资产时,才把具体使用到的资产复制进本仓库:
```text
assets/third_party/raw/<asset_type>/<category>/<asset_id>/
```
导入时必须记录:
- 原始 URL
- provider / site name
- license
- retrieval date
- source asset ID
- 本地路径
- 转换或清洗步骤
- 必要时记录 hash
候选第三方来源见:
```text
docs/external_source_catalog.md
docs/source_notes/third_party.md
```
重要规则:
- link-only candidate 不写入 `manifest/assets.jsonl`
- 下载数量不等于可生产资产数量
- 来源库总量不能直接相加,必须处理上游派生和重复资产
## 校验
在仓库根目录运行:
```bash
python scripts/validate_asset.py --all
```
当前期望输出:
```text
validated_assets=26
```
也可以校验单个资产:
```bash
python scripts/validate_asset.py assets/robocasa/raw/fixtures/dishwashers/Dishwasher031
```
校验内容包括:
- `metadata.yaml` 是否存在
- `entry_file` 是否存在
- 资产目录是否符合四层结构
- metadata 与目录字段是否一致
- metadata 与 manifest 是否一致
- MJCF 中的 mesh/texture 文件引用是否能解析
## 下载
```bash
export HF_ENDPOINT=https://hf-mirror.com
hf download yuchi233/Placement_Assets \
--repo-type dataset \
--local-dir ./Placement_Assets \
--max-workers 4
```
或使用脚本:
```bash
scripts/download_from_hf.sh yuchi233/Placement_Assets ./Placement_Assets
```
## 上传
先登录:
```bash
export HF_ENDPOINT=https://hf-mirror.com
hf auth login
```
上传:
```bash
cd /home/yuchi/asset-library
scripts/upload_to_hf.sh yuchi233/Placement_Assets
```
或者直接使用:
```bash
hf upload yuchi233/Placement_Assets . . --repo-type dataset
```
## 后续扩展流程
新增资产时按以下步骤:
1. 确认来源、license、可再分发性和用途。
2. 确定资产粒度:单资产还是 collection-level asset。
3. 放入统一四层目录结构。
4. 保留 upstream 文件名和相对路径,除非有兼容性原因。
5. 添加 `metadata.yaml`。
6. 更新 `manifest/assets.jsonl`。
7. 如果改了目录结构,更新 `docs/structure_mapping.md` 和 `docs/source_notes/<source>.md`。
8. 运行 `python scripts/validate_asset.py --all`。
9. 上传到 Hugging Face。
|