[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-1246":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":14,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":30,"readmeContent":31,"aiSummary":32,"trendingCount":16,"starSnapshotCount":16,"syncStatus":33,"lastSyncTime":34,"discoverSource":35},1246,"WorldEngine","OpenDriveLab\u002FWorldEngine","OpenDriveLab","WorldEngine: Towards the Era of Post-Training for Physical AI","https:\u002F\u002Fopendrivelab.com\u002FWorldEngine\u002F",null,"Python",337,18,15,1,0,5,12,49,3.84,"Apache License 2.0",false,"main",[25,26,27,28,29],"3dgs","end-to-end-autonomous-driving","post-training","simulation","world-model","2026-06-12 02:00:25","\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fimgs\u002FWE_title.png\" width=\"800px\">\n\n# Towards the Era of Post-Training for Physical AI\n> *The missing infrastructure for Physical AI post-training in AD. Open-source. Production-validated.*\n\n[![Paper](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaper-Coming_Soon-b31b1b.svg?style=for-the-badge&logo=arxiv)](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FWorldEngine)\n[![YouTube](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FYouTube-Video-FF0000.svg?style=for-the-badge&logo=youtube)](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=P1zEyfqa1uY)\n[![Hugging Face](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHugging_Face-Dataset-ffc107.svg?style=for-the-badge&logo=huggingface)](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenDriveLab\u002FWorldEngine)\n[![ModelScope](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FModelScope-Dataset-orange.svg?style=for-the-badge)](https:\u002F\u002Fwww.modelscope.cn\u002Fdatasets\u002FOpenDriveLab\u002FWorldEngine)\n\u003Cbr>\n[![PyTorch](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPyTorch-2.0.1-EE4C2C.svg?style=for-the-badge&logo=pytorch)](https:\u002F\u002Fpytorch.org)\n[![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fpython-3.9-blue?style=for-the-badge)](https:\u002F\u002Fwww.python.org)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-Apache_2.0-green.svg?style=for-the-badge)](https:\u002F\u002Fopensource.org\u002Flicenses\u002FApache-2.0)\n\n\u003C\u002Fdiv>\n\n\u003Cdiv id=\"top\" align=\"center\">\n\u003Cp align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fimgs\u002FREADME_overall.jpg\" width=\"800px\" >\n\u003C\u002Fp>\n\u003C\u002Fdiv>\n\n\n## Table of Contents\n- [Highlights](#highlights)\n- [News](#news)\n- [Benchmark](#benchmark)\n  - [Qualitative Results — Closed-Loop Simulation on nuPlan](#qualitative-results--closed-loop-simulation-on-nuplan)\n  - [On-Road Deployment — Night Urban Driving](#on-road-deployment--night-urban-driving)\n- [System Architecture](#system-architecture)\n- [Roadmap](#roadmap)\n- [Getting Started](#getting-started)\n  - [Documentation Overview](#documentation-overview)\n  - [Installation](#installation)\n  - [Quick Test](#quick-test)\n  - [Deep Dive by Module](#deep-dive-by-module)\n    - [SimEngine - Photorealistic Closed-Loop Simulation](#simengine---photorealistic-closed-loop-simulation)\n    - [AlgEngine - End-to-End Model Training \\& Fine-Tuning](#algengine---end-to-end-model-training--fine-tuning)\n    - [Scene Reconstruction - 3D Gaussian Splatting-based method, MTGS](#scene-reconstruction---3d-gaussian-splatting-based-method-mtgs)\n- [Citation](#citation)\n- [Contributing](#contributing)\n- [License](#license)\n- [Related Resources](#related-resources)\n\n## Highlights\n\n- **WorldEngine** is a post-training framework for Physical AI that systematically addresses the long-tail safety-critical data scarcity problem in autonomous driving.\n- **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.\n- **Photorealistic interactive simulation** via 3D Gaussian Splatting (3DGS): Each discovered scenario is reconstructed into a fully controllable, real-time-renderable simulation environment with independent dynamic agent manipulation.\n- **Behavior-driven scenario generation**: Leverages Behavior World Model (BWM) to generalize and synthesize diverse traffic variations from existing long-tail scenarios, expanding sparse safety-critical events into a dense, learnable distribution.\n- **RL-based post-training** on synthesized safety-critical rollouts substantially outperforms scaling pre-training data alone — competitive with a ~10× increase in pre-training data.\n- **Production-scale validation**: Deployed on a mass-produced ADAS platform trained on 80,000+ hours of real-world driving logs, reducing simulated collision rate by up to **45.5%** and achieving zero disengagements in a 200 km on-road test.\n\n\n## News\n- **[2026\u002F04\u002F09]** Official dataset released. See [OpenDriveLab\u002FWorldEngine](https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FOpenDriveLab\u002FWorldEngine) or [OpenDriveLab\u002FWorldEngine (ModelScope)](https:\u002F\u002Fwww.modelscope.cn\u002Fdatasets\u002FOpenDriveLab\u002FWorldEngine)\n- **[2026\u002F04\u002F10]** Official code repository established.\n\n\n## Benchmark\n\nWe compare different post-training paradigms on the nuPlan dataset, evaluating on both open-loop and closed-loop metrics across common and rare driving scenarios.\n\n> **Metric notes:**\n> **Early stage**. Stable ckpts and corresponding results coming soon.\n> - **Open-loop PDMS** is aligned with [NAVSIM v1.1](https:\u002F\u002Fgithub.com\u002Fautonomousvision\u002Fnavsim) PDM Score. *Common* denotes the standard `navtest` split; *Rare* denotes the `navtest_failures` subset — failure-prone rare-case scenarios extracted from `navtest`.\n> - **Closed-loop Success Rate** is defined as the fraction of simulated driving episodes completed without collision or off-road failure.\n> - **Closed-loop PDMS*** is the PDM Score obtained via **SimEngine closed-loop testing**, where the planner interacts with reactive agents in simulation under real-time rendering.\n>\n> **Training notes:**\n> - **Rare logs** are failure-prone scenarios automatically extracted from `navtrain` by the pre-trained agent itself (see [Rare Case Extraction](docs\u002Falgengine_usage.md#rare-case-extraction)). \n> - **Common logs** are the standard cases in `navtrain`.\n\n| Method | Open-loop PDMS ↑ (common) | Open-loop PDMS ↑ (rare) | Closed-loop Success Rate ↑ | Closed-loop PDMS* ↑ |\n|:-------|:-------------------------:|:-----------------------:|:--------------------------:|:--------------------:|\n| Base model | 85.62 | 47.15 | 73.61 | 60.28 |\n| Supervised fine-tuning on rare logs | 87.03 | 49.68 | 73.26 | 62.26 |\n| Post-training on common logs | 86.15 | 51.49 | 64.58 | 56.66 |\n| Post-training on rare logs | 89.29 | 62.56 | 74.31 | 62.55 |\n| Post-training on rare synthetic replays | 88.01 | 56.62 | 76.39 | 62.11 |\n| Post-training on rare rollouts w\u002Fo Behaviour WM | 88.99 | 59.69 | 85.07 | 68.29 |\n| **Post-training with WorldEngine** | **88.95** | **59.83** | **88.89** | **70.12** |\n\n**Key findings:**\n- Post-training on **rare logs** significantly outperforms supervised fine-tuning (62.56 vs 49.68 open-loop rare PDMS), demonstrating the advantage of reward-guided optimization over imitation.\n- Post-training on **common logs** provides limited benefit and even degrades closed-loop performance (success rate drops from 73.61% to 64.58%), confirming that long-tail event discovery is essential.\n- The full **WorldEngine** pipeline achieves the best closed-loop performance (**88.89%** success rate, **70.12** PDMS*), a **+15.28%** absolute improvement in success rate over the base model.\n\n### Qualitative Results — Closed-Loop Simulation on nuPlan\n\nEach pair shows the **Base model** vs **WorldEngine post-trained model** on the same rare-case scenario. Left: front-camera rendering; Right: BEV trajectory visualization.\n\n\u003Cdiv align=\"center\">\n\u003Ctable>\n\u003Ctr>\n\u003Ctd>\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fimgs\u002Fnuplan_1.png\" width=\"400px\">\u003C\u002Ftd>\n\u003Ctd>\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fimgs\u002Fnuplan_2.png\" width=\"400px\">\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fimgs\u002Fnuplan_3.png\" width=\"400px\">\u003C\u002Ftd>\n\u003Ctd>\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fimgs\u002Fnuplan_4.png\" width=\"400px\">\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\u003C\u002Fdiv>\n\n### On-Road Deployment — Night Urban Driving\n\nZero disengagements in 200 km on-road testing on a mass-produced ADAS platform.\n\n\u003Cdiv align=\"center\">\n\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fgif\u002FWE_road_night_01.gif\" width=\"270px\">\n\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fgif\u002FWE_road_night_02.gif\" width=\"270px\">\n\u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002FOpenDriveLab\u002Fopendrivelab.github.io\u002Frefs\u002Fheads\u002Fmaster\u002FWorldEngine\u002Fgif\u002FWE_road_night_03.gif\" width=\"270px\">\n\u003C\u002Fdiv>\n\n\n## System Architecture\n\nWorldEngine consists of two tightly coupled subsystems:\n\n\n| Module | Function | Core Technology |\n|--------|----------|----------------|\n| **[SimEngine](docs\u002FWE_simulation.md)** | Closed-loop simulation with ego & agents | Hydra, Ray, rendering |\n| **[AlgEngine](docs\u002FAlgEngine.md)** | End-to-end model training & evaluation | MMDetection3D, UniAD\u002FVADv2\u002FHydraMDP |\n\n\n## Roadmap\n\n- [x] Core platform integration (SimEngine + AlgEngine)\n- [x] Multi-GPU distributed simulation and training\n- [x] Rare case extraction and fine-tuning pipeline\n- [x] Comprehensive documentation and usage guides\n- [x] Hugging Face \u002F ModelScope dataset\n- [x] Open-source release (code, data, early pre-trained models)\n- [ ] arXiv preprint\n- [ ] Behavior World Model integration\n- [ ] Stable pre-trained models\n\n\n## Getting Started\n\n### Documentation Overview\n\nWorldEngine provides comprehensive guides for each stage of your workflow:\n\n| Guide | Purpose | Key Topics |\n|-------|---------|------------|\n| **[Installation](docs\u002Finstallation.md)** | Set up both conda environments | Two-environment setup (simengine + algengine), dependencies, troubleshooting |\n| **[Data Organization](docs\u002Fdata_organization.md)** | Prepare datasets and checkpoints | Data structure, Hugging Face\u002FModelScope downloads, symlinks |\n| **[Quick Start](docs\u002Fquick_start.md)** | Run your first experiment in 5 min | Quick test tutorial, understanding results, complete pipeline |\n| **[SimEngine Usage](docs\u002Fsimengine_usage.md)** | Master closed-loop simulation | Rollout scripts, distributed testing, configuration, metrics |\n| **[AlgEngine Usage](docs\u002Falgengine_usage.md)** | Train and fine-tune models | Training from scratch, evaluation, rare case extraction, RL fine-tuning |\n\n### Installation\n\nWorldEngine requires **two separate conda environments** due to different Python requirements.\n\n**Full installation guide:** [docs\u002Finstallation.md](docs\u002Finstallation.md)\n\n### Quick Test\n\nVerify your installation with a pre-trained model:\n\n```bash\n# Set up environment variable\nexport WORLDENGINE_ROOT=$(pwd)\n\n# Option 1: Single GPU test \nbash scripts\u002Fclosed_loop_test.sh\n\n# Option 2: Multi-GPU test (Default 8 GPUs)\nbash scripts\u002Fmultigpu_closed_loop_test.sh\n```\n\n**What this does:**\n- Loads a pre-trained VADv2 model (50% training data, epoch 8)\n- Runs closed-loop simulation on 288 rare-case test scenarios\n- Evaluates with navsim v1 PDMS (collision avoidance, progress, comfort, etc.)\n- Saves results to `experiments\u002Fclosed_loop_exps\u002Fe2e_vadv2_50pct\u002Fnavtest_failures_NR\u002F`\n\n**Detailed quick start tutorial:** [docs\u002Fquick_start.md](docs\u002Fquick_start.md)\n\n### Deep Dive by Module\n\nAfter the quick test, explore each subsystem in detail:\n\n#### SimEngine - Photorealistic Closed-Loop Simulation\n\nLearn how to run simulations, generate rollouts, and test models:\n\n- **Rollout scripts** for data generation (no model required)\n- **Testing scripts** for model evaluation (single\u002Fmulti-GPU)\n- **Ray distributed simulation** for large-scale testing\n- **Reactive vs non-reactive** agent modes\n- **Configuration guide** for all Hydra parameters\n\n**[SimEngine Usage Guide](docs\u002Fsimengine_usage.md)**\n\n#### AlgEngine - End-to-End Model Training & Fine-Tuning\n\nLearn how to train models, extract rare cases, and fine-tune:\n\n- **Training from scratch**\n- **Open-loop evaluation** on test sets\n- **Rare case extraction** from evaluation failures\n- **RL-based fine-tuning** on long-tail scenarios\n- **Multi-GPU training** with distributed data parallel\n\n**[AlgEngine Usage Guide](docs\u002Falgengine_usage.md)**\n\n#### Scene Reconstruction - 3D Gaussian Splatting-based method, MTGS\n\nWorldEngine's simulation environments are powered by 3D Gaussian Splatting (MTGS):\n\n- **Multi-traversal reconstruction** from nuPlan data\n- **Photorealistic rendering** for closed-loop simulation\n- **Asset generation** for SimEngine scenes\n\n**[MTGS Repository](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FMTGS)**\n\n\n## Citation\n\nIf any parts of our work help your research, please consider citing us and giving a star to our repository:\n\nIf you use the Render Assets (MTGS), please also cite:\n```bibtex\n@article{li2025mtgs,\n  title={MTGS: Multi-Traversal Gaussian Splatting},\n  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},\n  journal={arXiv preprint arXiv:2503.12552},\n  year={2025}\n}\n```\nIf you use the augmented scenarios data, please cite as well:\n```bibtex\n@inproceedings{zhou2025nexus,\n  title={Decoupled Diffusion Sparks Adaptive Scene Generation},\n  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},\n  booktitle={ICCV},\n  year={2025}\n}\n```\n```bibtex\n@article{li2025optimization,\n  title={Optimization-Guided Diffusion for Interactive Scene Generation},\n  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},\n  journal={arXiv preprint arXiv:2512.07661},\n  year={2025}\n}\n```\nIf you find AlgEngine well, please cite as well:\n```bibtex\n@ARTICLE{11353028,\n  author={Liu, Haochen and Li, Tianyu and Yang, Haohan and Chen, Li and Wang, Caojun and Guo, Ke and Tian, Haochen and Li, Hongchen and Li, Hongyang and Lv, Chen},\n  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, \n  title={Reinforced Refinement With Self-Aware Expansion for End-to-End Autonomous Driving}, \n  year={2026},\n  volume={48},\n  number={5},\n  pages={5774-5792},\n  keywords={Adaptation models;Self-aware;Autonomous vehicles;Pipelines;Planning;Training;Reinforcement learning;Uncertainty;Data models;Safety;End-to-end autonomous driving;reinforced finetuning;imitation learning;motion planning},\n  doi={10.1109\u002FTPAMI.2026.3653866}}\n```\nIf you find data scaling infos helpful, please also cite:\n```bibtex\n@article{tian2025simscale,\n        title={SimScale: Learning to Drive via Real-World Simulation at Scale},\n        author={Haochen Tian and Tianyu Li and Haochen Liu and Jiazhi Yang and Yihang Qiu and Guang Li and Junli Wang and Yinfeng Gao and Zhang Zhang and Liang Wang and Hangjun Ye and Tieniu Tan and Long Chen and Hongyang Li},\n        journal={arXiv preprint arXiv:2511.23369},\n        year={2025}\n      }\n```\n\n## Contributing\n\nWe welcome contributions from the community! Whether you want to:\n\n- **Report bugs** - Open an [Issue](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FWorldEngine\u002Fissues)\n- **Improve documentation** - Submit a [Pull Request](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FWorldEngine\u002Fpulls)\n- **Contribute code** - Fork, develop, and submit a PR\n\nPlease read our contributing guidelines before submitting PRs.\n\n**For questions:**\n1. Check the [documentation](docs\u002F) first\n2. Search existing [Issues](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FWorldEngine\u002Fissues)\n\n\n## License\n\nAll content in this repository is under the [Apache-2.0 license](https:\u002F\u002Fwww.apache.org\u002Flicenses\u002FLICENSE-2.0).\n\nThe released data is based on [nuPlan](https:\u002F\u002Fwww.nuscenes.org\u002Fnuplan) and is under the [CC-BY-NC-SA 4.0](https:\u002F\u002Fcreativecommons.org\u002Flicenses\u002Fby-nc-sa\u002F4.0\u002F) license.\n\n\n\u003C!-- ## 👥 Team & Acknowledgements\n\nWorldEngine is developed by **Shanghai Innovation Institute (SII)** and **OpenDriveLab** at The University of Hong Kong.\n\n**Core Contributors:**\n- Scene Reconstruction Team\n- Simulation Platform Team\n- Algorithm Development Team\n\nWe would like to thank all contributors and the open-source community for their support. -->\n\n\n## Related Resources\n\nWe acknowledge all the open-source contributors for the following projects to make this work possible:\n\n\u003Cdiv align=\"center\">\n\n| Project | Description |\n|:-------:|:------------|\n| [![MTGS](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMTGS-Multi--Traversal_GS-blue?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FMTGS) | Multi-traversal Gaussian Splatting for scene reconstruction |\n| [![SimScale](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FSimScale-Driving_Simulation-AD9BC2?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FSimScale) | Large scale driving simulation |\n| [![nerfstudio](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002Fnerfstudio-NeRF_Framework-green?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002Fnerfstudio-project\u002Fnerfstudio) | Collaboration-friendly NeRF toolkit |\n| [![MMDetection3D](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMMDetection3D-3D_Detection-orange?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002Fopen-mmlab\u002Fmmdetection3d) | 3D detection framework |\n| [![UniAD](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FUniAD-End--to--End_AD-red?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FUniAD) | End-to-end autonomous driving framework |\n| [![VADv2](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FVADv2-End--to--End_AD-crimson?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002Fpriest-yang\u002FVADv2) | Vectorized autonomous driving framework |\n| [![NAVSIM](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNAVSIM-AD_Benchmark-teal?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002Fautonomousvision\u002Fnavsim) | Non-reactive autonomous vehicle simulation benchmark |\n| [![nuPlan](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FnuPlan-Dataset-purple?style=flat-square&logo=github)](https:\u002F\u002Fwww.nuscenes.org\u002Fnuplan) | Large-scale autonomous driving dataset |\n| [![MetaDrive](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FMetaDrive-Driving_Simulation-ff69b4?style=flat-square&logo=github)](https:\u002F\u002Fgithub.com\u002Fmetadriverse\u002Fmetadrive) | Compositional driving simulation platform |\n| [![Ray](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FRay-Distributed_Computing-yellow?style=flat-square&logo=ray)](https:\u002F\u002Fgithub.com\u002Fray-project\u002Fray) | Distributed execution framework |\n| [![Hydra](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHydra-Config_Management-lightblue?style=flat-square&logo=python)](https:\u002F\u002Fgithub.com\u002Ffacebookresearch\u002Fhydra) | Configuration management framework |\n\n\u003C\u002Fdiv>\n\n---\n\n\u003Cdiv align=\"center\">\n\n**If you find WorldEngine useful, please consider giving us a star!**\n\n**Quick Links:** [Documentation](docs\u002F) | [Installation](docs\u002Finstallation.md) | [Quick Start](docs\u002Fquick_start.md) | [Issues](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FWorldEngine\u002Fissues) | [Discussions](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FWorldEngine\u002Fdiscussions)\n\n**Contact:** For research collaboration or questions, visit our [Discussions](https:\u002F\u002Fgithub.com\u002FOpenDriveLab\u002FWorldEngine\u002Fdiscussions)\n\n\u003C\u002Fdiv>\n","WorldEngine是一个面向物理AI的后训练框架，旨在解决自动驾驶中长尾安全关键数据稀缺的问题。其核心功能包括通过预训练代理自动识别真实驾驶日志中的高风险场景，无需人工设计或合成扰动；提供基于3D高斯点云技术的场景重建方法MTGS，以及支持光真级闭环模拟和端到端模型训练与微调的SimEngine与AlgEngine模块。该项目适合于需要提高自动驾驶系统在复杂、罕见情况下的鲁棒性和安全性场景使用，特别适用于研发阶段对算法进行验证和优化。采用Python编写，并已通过生产环境验证。",2,"2026-06-11 02:42:34","CREATED_QUERY"]