Datasets:

Languages:
English
ArXiv:
License:
OpenDriveLab-org commited on
Commit
4533c44
·
1 Parent(s): 2bfed32

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +400 -4
README.md CHANGED
@@ -1,7 +1,403 @@
1
  ---
2
  license: cc-by-nc-sa-4.0
3
- task_categories:
4
- - robotics
5
  language:
6
- - en
7
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
  license: cc-by-nc-sa-4.0
 
 
3
  language:
4
+ - en
5
+ tags:
6
+ - Robotics
7
+ ---
8
+ <div align="center">
9
+ <img src="docs/imgs/WE_title.png" width="800px">
10
+
11
+ # **WorldEngine: Towards the Era of Post-Training for Physical AI**
12
+
13
+ [![License](https://img.shields.io/badge/License-CC--BY--NC--SA--4.0-blue.svg?style=for-the-badge)](https://creativecommons.org/licenses/by-nc-sa/4.0/)
14
+ [![ModelScope](https://img.shields.io/badge/ModelScope-Dataset-orange.svg?style=for-the-badge)](https://www.modelscope.cn/datasets/OpenDriveLab/WorldEngine)
15
+ [![GitHub](https://img.shields.io/badge/GitHub-Repository-black.svg?style=for-the-badge&logo=github)](https://github.com/OpenDriveLab/WorldEngine)
16
+ [![Hugging Face](https://img.shields.io/badge/Hugging_Face-Dataset-ffc107.svg?style=for-the-badge&logo=huggingface)](https://huggingface.co/datasets/OpenDriveLab/WorldEngine)
17
+
18
+ </div>
19
+
20
+ > **WorldEngine** is the first post-training framework for Physical AI. This dataset contains preprocessed data, model checkpoints and reconstruction assets for algorithm training and closed-loop simulation.
21
+
22
+ <p align="center">
23
+ <img src="docs/imgs/README_overall.png" width="800px" >
24
+ </p>
25
+
26
+ > Joint effort by OpenDriveLab at The University of Hong Kong, Huawei Inc. and Shanghai Innovation Institute (SII).
27
+
28
+ ## Highlights
29
+
30
+ - **First post-training framework for Physical AI**: Systematically addresses the long-tail safety-critical data scarcity problem in autonomous driving.
31
+ - **Data-driven long-tail discovery**: Failure-prone scenarios are automatically identified from real-world driving logs by the pre-trained agent itself — no manual design, no synthetic perturbations.
32
+ - **Photorealistic interactive simulation** via 3D Gaussian Splatting (3DGS): Each discovered scenario is reconstructed into a fully controllable, real-time-renderable simulation environment.
33
+ - **Behavior-driven scenario generation**: Leverages Behavior World Model (BWM) to generalize and synthesize diverse traffic variations from long-tail scenarios, expanding sparse safety-critical events into a dense, learnable distribution.
34
+ - **RL-based post-training** on safety-critical rollouts substantially outperforms scaling pre-training data alone — competitive with a ~10x increase in pre-training data.
35
+ - **Production-scale validation**: Deployed on a mass-produced ADAS platform trained on 80,000+ hours of driving logs, reducing collision rate by up to **45.5%** and achieving zero disengagements in a 200 km on-road test.
36
+
37
+ ## News
38
+
39
+ - **[2026/04/09]** Official code repository established. Data publication under preparation.
40
+ - **[2026/04/09]** Official data release.
41
+
42
+ ---
43
+
44
+ ## Table of Contents
45
+
46
+ - [Highlights](#highlights)
47
+ - [News](#news)
48
+ - [Dataset Overview](#-dataset-overview)
49
+ - [Download](#-download)
50
+ - [Directory Structure](#-directory-structure)
51
+ - [Benchmark](#-benchmark)
52
+ - [Environment Setup](#️-environment-setup)
53
+ - [Usage](#-usage)
54
+ - [Citation](#-citation)
55
+ - [License](#-license)
56
+ - [Related Links](#-related-links)
57
+ - [Contact](#-contact)
58
+
59
+ ## 📦 Dataset Overview
60
+
61
+ This dataset uses a **modular data structure** where each subsystem (AlgEngine, SimEngine) has its own data requirements while sharing common formats.
62
+
63
+ | Module | Function | Data Types |
64
+ |--------|----------|-----------|
65
+ | **Raw Data** | nuPlan & OpenScene base datasets | Sensor data, maps, annotations |
66
+ | **AlgEngine** | End-to-end model training & evaluation | Preprocessed annotations, ckpts, caches |
67
+ | **SimEngine** | Closed-loop simulation environments | Scene assets, config files |
68
+
69
+ ```bash
70
+ WorldEngine/
71
+ └── data/ # Main data directory
72
+ ├── raw/ # Raw datasets (nuPlan, OpenScene)
73
+ ├── alg_engine/ # AlgEngine-specific data
74
+ └── sim_engine/ # SimEngine-specific data
75
+ ```
76
+
77
+ ## 🚀 Download
78
+
79
+ ### Option 1: ModelScope SDK
80
+
81
+ Recommended for quick download:
82
+ ```
83
+ from modelscope import MsDataset
84
+
85
+ # Load the dataset
86
+ ds = MsDataset.load('OpenDriveLab/WorldEngine')
87
+ ```
88
+
89
+ ### Option 2: Git Clone
90
+
91
+ For full version control:
92
+ ```
93
+ # Install Git LFS (Large File Storage)
94
+ git lfs install
95
+
96
+ git clone https://www.modelscope.cn/datasets/OpenDriveLab/WorldEngine.git
97
+ ```
98
+
99
+ ### Option 3: ModelScope CLI
100
+ ```
101
+ pip install modelscope
102
+ modelscope download --dataset OpenDriveLab/WorldEngine
103
+ ```
104
+
105
+ ---
106
+
107
+ ## 📂 Directory Structure
108
+
109
+ ### 1️⃣ Raw Data (`data/raw/`)
110
+
111
+ <details>
112
+ <summary><b>Click to expand full directory structure</b></summary>
113
+
114
+ After downloading the **nuPlan** and **OpenScene** raw datasets, set up the following structure via symlinks (`ln -s`):
115
+
116
+ ```bash
117
+ data/raw/
118
+ ├── nuplan/ # nuPlan raw dataset
119
+ │ ├── maps/ # HD maps (required by all modules)
120
+ │ │ ├── us-nv-las-vegas-strip/
121
+ │ │ ├── us-ma-boston/
122
+ │ │ ├── us-pa-pittsburgh-hazelwood/
123
+ │ │ └── sg-one-north/
124
+ │ ├── sensor_blobs/ # Camera images and LiDAR
125
+ │ └── splits/ # Train/val/test splits
126
+
127
+ └── openscene-v1.1/ # OpenScene dataset (based on nuPlan)
128
+ ├── sensor_blobs/
129
+ │ ├── trainval/ # Training sensor data
130
+ │ └── test/ # Test sensor data
131
+ └── meta_datas/
132
+ ├── trainval/ # Training metadata
133
+ └── test/ # Test metadata
134
+ ```
135
+
136
+ </details>
137
+
138
+ ### 2️⃣ AlgEngine Data (`data/alg_engine/`)
139
+
140
+ <details>
141
+ <summary><b>Click to expand full directory structure</b></summary>
142
+
143
+ Data for **end-to-end model training and evaluation**:
144
+
145
+ ```bash
146
+ data/alg_engine/
147
+ ├── openscene-synthetic/ # Synthetic data generated by SimEngine (need to generate)
148
+ │ ├── sensor_blobs/
149
+ │ ├── meta_datas/
150
+ │ └── pdms_pkl/
151
+
152
+ ├── ckpts/ # Pre-trained model checkpoints
153
+ │ ├── bevformerv2-r50-t1-base_epoch_48.pth
154
+ │ ├── e2e_vadv2_50pct_ep8.pth
155
+ │ ├── e2e_vadv2_50pct_rlft_synthetic_ep8.pth
156
+ │ ├── e2e_vadv2_50pct_rlft_synthetic_ep8_merged.pth
157
+ │ ├── track_map_nuplan_r50_navtrain_100pct_bs1x8.pth
158
+ │ └── track_map_nuplan_r50_navtrain_50pct_bs1x8.pth
159
+
160
+ ├── pdms_cache/ # Pre-computed PDM metric caches
161
+ │ ├── pdm_8192_gt_cache_navtest.pkl
162
+ │ └── pdm_8192_gt_cache_navtrain.pkl
163
+
164
+ ├── merged_infos_navformer/ # Preprocessed annotations
165
+ │ ├── nuplan_openscene_navtest.pkl
166
+ │ └── nuplan_openscene_navtrain.pkl
167
+
168
+ └── test_8192_kmeans.npy # K-means clustering for PDM
169
+ ```
170
+
171
+ </details>
172
+
173
+ ### 3️⃣ SimEngine Data (`data/sim_engine/`)
174
+
175
+ <details>
176
+ <summary><b>Click to expand full directory structure</b></summary>
177
+
178
+ Data for **closed-loop simulation**:
179
+
180
+ ```bash
181
+ data/sim_engine/
182
+ ├── assets/ # Simulation scene assets (need extraction)
183
+ │ ├── navtest/ # navtest scene assets (10 parts)
184
+ │ ├── navtrain/ # navtrain scene assets (82 parts)
185
+ │ └── navtest_failures/ # navtest rare logs scene assets
186
+
187
+ └── scenarios/ # Scenario configurations
188
+ ├── original/ # Original logged scenarios
189
+ │ ├── navtest_failures/
190
+ │ ├── navtrain_50pct_collision/
191
+ │ ├── navtrain_ep_per1/
192
+ │ ├── navtrain_failures_per1/
193
+ │ └── navtrain_hydramdp_failures/
194
+ └── augmented/ # Augmented scenarios (from BWM)
195
+ ├── navtrain_50pct_collision/
196
+ ├── navtrain_50pct_ep_1pct/
197
+ └── navtrain_50pct_offroad/
198
+ ```
199
+
200
+ **⚠️ Important: Scene Asset Extraction**
201
+
202
+ Scene assets in the `assets/` directory are stored as split archives and must be extracted before use:
203
+
204
+ ```bash
205
+ cd data/sim_engine/assets
206
+
207
+ # Extract navtest scene assets (10 parts)
208
+ cd navtest
209
+ cat navtest.tar.gz.part* > navtest.tar.gz
210
+ tar -xzf navtest.tar.gz --strip-components=1 # Remove top-level directory from archive
211
+ rm navtest.tar.gz # Optional: remove merged archive to save space
212
+
213
+ # Extract navtrain scene assets (82 parts)
214
+ cd ../navtrain
215
+ cat navtrain.tar.gz.part* > navtrain.tar.gz
216
+ tar -xzf navtrain.tar.gz --strip-components=1
217
+ rm navtrain.tar.gz
218
+
219
+ # Extract navtest_failures scene assets
220
+ cd ../navtest_failures
221
+ cat navtest_failures.tar.gz.part* > navtest_failures.tar.gz
222
+ tar -xzf navtest_failures.tar.gz --strip-components=1
223
+ rm navtest_failures.tar.gz
224
+
225
+ cd ../../.. # Return to WorldEngine root
226
+ ```
227
+
228
+ 💡 **Tips**:
229
+ - The `--strip-components=1` parameter ensures extraction to the current directory, avoiding nested structures like `navtest/navtest/`
230
+ - Extracted scene assets contain all files needed for 3D Gaussian Splatting (3DGS) rendering; each scene is approximately several hundred MB
231
+
232
+ </details>
233
+
234
+ ---
235
+
236
+ ## ⚙️ Environment Setup
237
+
238
+ Configure the following environment variables for proper data access:
239
+
240
+ ### Quick Configuration
241
+
242
+ ```bash
243
+ # Add to ~/.bashrc or ~/.zshrc
244
+ export WORLDENGINE_ROOT="/path/to/WorldEngine"
245
+ export NUPLAN_MAPS_ROOT="${WORLDENGINE_ROOT}/data/raw/nuplan/maps"
246
+ export PYTHONPATH=$WORLDENGINE_ROOT:$PYTHONPATH
247
+ ```
248
+
249
+ ### Apply Changes
250
+
251
+ ```bash
252
+ source ~/.bashrc # or source ~/.zshrc
253
+ ```
254
+
255
+ 💡 **Tip**: After adding the above to your shell config file, these environment variables will be automatically loaded every time you open a new terminal.
256
+
257
+ ---
258
+
259
+ ## 📖 Usage
260
+
261
+ ### Quick Start
262
+
263
+ Follow these steps to set up the dataset:
264
+
265
+ | Step | Action | Description |
266
+ |:----:|--------|-------------|
267
+ | **1** | Download dataset | Use ModelScope SDK or Git Clone |
268
+ | **2** | Extract scene assets | Extract split archives in `data/sim_engine/assets/` ([see instructions](#3️⃣-simengine-data-datasim_engine)) |
269
+ | **3** | Set environment variables | Configure `WORLDENGINE_ROOT` and related paths |
270
+ | **4** | Create symlinks | Link raw datasets (if needed) |
271
+ | **5** | Verify installation | Run the quick test script |
272
+
273
+ ### Detailed Setup
274
+
275
+ <details>
276
+ <summary><b>2. Extract Scene Assets (Required)</b></summary>
277
+
278
+ SimEngine scene assets are stored as split archives and must be extracted before use:
279
+
280
+ ```bash
281
+ cd data/sim_engine/assets
282
+
283
+ # Extract all scene assets
284
+ for dir in navtest navtrain navtest_failures; do
285
+ echo "Processing ${dir}..."
286
+ cd ${dir}
287
+ cat ${dir}.tar.gz.part* > ${dir}.tar.gz
288
+ tar -xzf ${dir}.tar.gz --strip-components=1 # Avoid nested directories
289
+ rm ${dir}.tar.gz # Optional: remove merged archive
290
+ cd ..
291
+ done
292
+
293
+ cd ../../..
294
+ ```
295
+
296
+ Or extract them manually one by one ([see detailed instructions in SimEngine Data section](#3️⃣-simengine-data-datasim_engine)).
297
+
298
+ </details>
299
+
300
+ <details>
301
+ <summary><b>4. Create Symlinks (Optional)</b></summary>
302
+
303
+ If you have already downloaded nuPlan and OpenScene data, use symlinks to avoid data duplication:
304
+
305
+ ```bash
306
+ cd WorldEngine/data/raw
307
+ ln -s /path/to/nuplan nuplan
308
+ ln -s /path/to/openscene-v1.1 openscene-v1.1
309
+ cd openscene-v1.1
310
+ ln -s ../nuplan/maps maps
311
+ ```
312
+
313
+ </details>
314
+
315
+ ### Next Steps
316
+
317
+ After dataset setup, refer to the main project documentation:
318
+
319
+ - 📘 [Installation Guide](https://github.com/OpenDriveLab/WorldEngine/blob/main/docs/installation.md)
320
+ - 🚀 [Quick Start](https://github.com/OpenDriveLab/WorldEngine/blob/main/docs/quick_start.md)
321
+ - 🎮 [SimEngine Usage Guide](https://github.com/OpenDriveLab/WorldEngine/blob/main/docs/simengine_usage.md)
322
+ - 🧠 [AlgEngine Usage Guide](https://github.com/OpenDriveLab/WorldEngine/blob/main/docs/algengine_usage.md)
323
+
324
+ ---
325
+
326
+ ## 📝 Citation
327
+
328
+ If this project is helpful to your research, please consider citing:
329
+
330
+ ```bibtex
331
+
332
+ ```
333
+
334
+ If you use the Render Assets (MTGS), please also cite:
335
+ ```bibtex
336
+ @article{li2025mtgs,
337
+ title={MTGS: Multi-Traversal Gaussian Splatting},
338
+ author={Li, Tianyu and Qiu, Yihang and Wu, Zhenhua and Lindstr{\"o}m, Carl and Su, Peng and Nie{\ss}ner, Matthias and Li, Hongyang},
339
+ journal={arXiv preprint arXiv:2503.12552},
340
+ year={2025}
341
+ }
342
+ ```
343
+
344
+ If you use the scenario data generated by Behavior World Model (BWM), please also cite:
345
+ ```bibtex
346
+ @inproceedings{zhou2025nexus,
347
+ title={Decoupled Diffusion Sparks Adaptive Scene Generation},
348
+ author={Zhou, Yunsong and Ye, Naisheng and Ljungbergh, William and Li, Tianyu and Yang, Jiazhi and Yang, Zetong and Zhu, Hongzi and Petersson, Christoffer and Li, Hongyang},
349
+ booktitle={ICCV},
350
+ year={2025}
351
+ }
352
+ ```
353
+ ```bibtex
354
+ @article{li2025optimization,
355
+ title={Optimization-Guided Diffusion for Interactive Scene Generation},
356
+ author={Li, Shihao and Ye, Naisheng and Li, Tianyu and Chitta, Kashyap and An, Tuo and Su, Peng and Wang, Boyang and Liu, Haiou and Lv, Chen and Li, Hongyang},
357
+ journal={arXiv preprint arXiv:2512.07661},
358
+ year={2025}
359
+ }
360
+ ```
361
+
362
+ ---
363
+
364
+ ## 📄 License
365
+
366
+ This dataset is released under the **[CC-BY-NC-SA-4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/)** license.
367
+
368
+ ### Terms of Use
369
+
370
+ - ✅ **Allowed**: Modification, distribution, private use
371
+ - 📝 **Required**: Attribution, share alike
372
+ - ⚠️ **Restricted**: No commercial use; copyright and license notices must be retained
373
+
374
+ ---
375
+
376
+ ## 🔗 Related Links
377
+
378
+ | Resource | Link |
379
+ |:--------:|:-----|
380
+ | 🏠 **Project Home** | [WorldEngine GitHub](https://github.com/OpenDriveLab/WorldEngine) |
381
+ | 📦 **ModelScope** | [Dataset Page](https://www.modelscope.cn/datasets/OpenDriveLab/WorldEngine) |
382
+ | 💬 **Feedback** | [ModelScope Issues](https://www.modelscope.cn/datasets/OpenDriveLab/WorldEngine/feedback) |
383
+ | 📖 **Full Documentation** | [Documentation](https://github.com/OpenDriveLab/WorldEngine/tree/main/docs) |
384
+ | 🎨 **Scene Reconstruction** | [MTGS Repository](https://github.com/OpenDriveLab/MTGS) |
385
+
386
+ ---
387
+
388
+ ## 📧 Contact
389
+
390
+ For questions or suggestions, feel free to reach out:
391
+
392
+ - 🐛 **Bug Reports**: [GitHub Issues](https://github.com/OpenDriveLab/WorldEngine/issues)
393
+ - 📮 **Dataset Feedback**: [ModelScope Feedback](https://www.modelscope.cn/datasets/OpenDriveLab/WorldEngine/feedback)
394
+
395
+ ---
396
+
397
+ <div align="center">
398
+
399
+ **⭐ If you find WorldEngine useful, please consider giving us a Star! ⭐**
400
+
401
+ **Thank you for your support of the WorldEngine project!**
402
+
403
+ </div>