[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-83554":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":17,"stars30d":17,"stars90d":16,"forks30d":16,"starsTrendScore":18,"compositeScore":19,"rankGlobal":10,"rankLanguage":10,"license":20,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":23,"hasPages":21,"topics":24,"createdAt":10,"pushedAt":10,"updatedAt":25,"readmeContent":26,"aiSummary":10,"trendingCount":16,"starSnapshotCount":16,"syncStatus":27,"lastSyncTime":28,"discoverSource":29},83554,"boundless-world-model","boundless-large-model\u002Fboundless-world-model","boundless-large-model","High-fidelity world models for general embodied intelligence, such as data engines and world  simulators.","",null,"Python",1829,70,15,1,0,35,105,18.55,"Apache License 2.0",false,"main",true,[],"2026-06-12 02:04:34","\u003Cdiv align=\"center\">\n\n\u003Ch1>🌍 Boundless-World-Model \u003C\u002Fh1>\n\n\u003Cp align=\"center\">\n    \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FWorldArena\u002FWorldArena\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🏆_Leaderboard-WorldArena-yellow?style=flat\">\u003C\u002Fa>  \n    \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FBLM-Lab\u002FBoundless-World-Model\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F🤗_Model-BWM-blue?style=flat\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n\u003C\u002Fdiv>\n\n> **BWM** is a physically consistent, action-conditioned video world model built upon Wan2.2-TI2V-5B, serving as a low-cost yet high-fidelity simulator for robotic manipulation.\n\n## 🗞️ News\n\n- **[2026-05]** 🏆 **Top results on WorldArena Leaderboard!** BLM ranks 1st among open-source models on Track 1 and Track 2 Data Engine, while BWM-fast ranks 2nd overall on Track 1.\n- **[2026-05]** 🚀 **Inference code released!** Generate action-conditioned robot manipulation videos with BWM. See [🛠️ Usage](#️-usage).\n- **[2026-05]** 🎉 **Model definition released!** The BWM architecture and core model components are now available.\n\n## 🏆 Competition Results\n\n### **CVPR 2026 WorldArena Challenge**\n\n- **BLM**: 🥇 **1st Place** among open-source models on **Track 1** and **Track 2 Data Engine**.\n- **BWM-fast**: 🥈 **2nd Place** on the overall **Track 1** leaderboard.\n\n\u003Ctable align=\"center\">\n  \u003Ctr>\n    \u003Ctd align=\"center\">\n      \u003Cimg src=\"assets\u002Fimages\u002Ftrack-1-open-source.png\" alt=\"Track 1 open-source leaderboard\" width=\"800\">\u003Cbr>\n      \u003Csub>Track 1 open-source leaderboard\u003C\u002Fsub>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd align=\"center\">\n      \u003Cimg src=\"assets\u002Fimages\u002Ftrack-2-DE-open-source.png\" alt=\"Track 2 Data Engine open-source leaderboard\" width=\"800\">\u003Cbr>\n      \u003Csub>Track 2 Data Engine open-source leaderboard\u003C\u002Fsub>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n  \u003Ctr>\n    \u003Ctd align=\"center\">\n      \u003Cimg src=\"assets\u002Fimages\u002Ftrack-1-total.png\" alt=\"Track 1 overall leaderboard\" width=\"800\">\u003Cbr>\n      \u003Csub>Track 1 overall leaderboard\u003C\u002Fsub>\n    \u003C\u002Ftd>\n  \u003C\u002Ftr>\n\u003C\u002Ftable>\n\nLeaderboard: https:\u002F\u002Fhuggingface.co\u002Fspaces\u002FWorldArena\u002FWorldArena\n\n## Table of Contents\n- [✅ TODO](#-todo)\n- [🏗️ Framework](#️-framework)\n- [🎬 Qualitative Results](#-qualitative-results)\n- [🛠️ Usage](#️-usage)\n- [🏋️ Training](#️-training)\n- [🙏 Acknowledgements](#-acknowledgements)\n- [📜 Citing](#-Citing)\n\n---\n\n## ✅ TODO\n\n- [x] Release inference code\n- [x] Release model definition\n- [x] Release model weights\n- [ ] Release training code\n- [ ] Release technical report\n\n---\n\n## 🏗️ Framework\n\nComing soon !\n\n---\n\n## 🎬 Qualitative Results\n\n### **CVPR 2026 WorldArena Challenge**\n\n> The following simulation scenes are generated autoregressively by **BWM** from initial frames and action sequences in the [**WorldArena test set**](https:\u002F\u002Fgithub.com\u002Ftsinghua-fib-lab\u002FWorldArena\u002F), achieving high-fidelity visual realism while maintaining long-horizon physical consistency.\n\n#### 🧩 Scene 1: Compositional Spatial Rearrangement\n\n  \u003Ctable align=\"center\" >\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fblocks_ranking_size\u002Fepisode228.gif\" alt=\"blocks ranking size\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fstack_bowls_three\u002Fepisode152.gif\" alt=\"stack bowls three\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n  \u003C\u002Ftable>\n\n- **Task**: arrange blocks by size, stack bowls\n- **Challenge**: Multi-object spatial ordering, stacking stability, and contact-rich placement\n- **Ours**:\n  - ✅ Preserves object identity and target layout\n  - ✅ Maintains stable stacking contacts\n  - ✅ Predicts adaptive gripper control\n\n#### 🚪 Scene 2: Articulated Hinge Interaction\n\n  \u003Ctable align=\"center\" >\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fopen_microwave\u002Fepisode347.gif\" alt=\"open microwave\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fopen_laptop\u002Fepisode330.gif\" alt=\"open laptop\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n  \u003C\u002Ftable>\n\n- **Task**: open microwave, open laptop\n- **Challenge**: Articulated hinge motion, constrained rotation, and persistent object state\n- **Ours**:\n  - ✅ Captures hinge-constrained opening dynamics\n  - ✅ Maintains coherent object geometry during rotation\n  - ✅ Preserves opened states over long-horizon rollouts\n\n#### 🕹️ Scene 3: Fine-Grained Affordance Interaction\n\n  \u003Ctable align=\"center\" >\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fturn_switch\u002Fepisode674.gif\" alt=\"turn switch\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fhanging_mug\u002Fepisode373.gif\" alt=\"hanging mug\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fclick_bell\u002Fepisode796.gif\" alt=\"click bell\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fstamp_seal\u002Fepisode581.gif\" alt=\"stamp seal\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n  \u003C\u002Ftable>\n\n- **Task**: turn switch, hang mug, click bell, stamp seal\n- **Challenge**: Small contact regions, constrained placement, and precise state-changing interactions\n- **Ours**:\n  - ✅ Captures fine-grained affordance dynamics\n  - ✅ Aligns contact with object affordances\n  - ✅ Preserves state-changing interactions\n\n#### 🤝 Scene 4: Bimanual Coordination and Handover\n\n  \u003Ctable align=\"center\" >\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fhandover_block\u002Fepisode47.gif\" alt=\"handover block\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fhandover_mic\u002Fepisode298.gif\" alt=\"handover mic\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n  \u003C\u002Ftable>\n\n- **Task**: hand over block, hand over mic\n- **Challenge**: Dual-arm synchronization, inter-arm occlusion, and coordinated grasp timing\n- **Ours**:\n  - ✅ Models synchronized dual-arm motion\n  - ✅ Preserves object continuity\n  - ✅ Avoids close-contact collisions\n\n#### 📦 Scene 5: Long-Horizon Constrained Placement\n\n  \u003Ctable align=\"center\" >\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fput_object_cabinet\u002Fepisode33.gif\" alt=\"put object cabinet\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fput_bottles_dustbin\u002Fepisode1.gif\" alt=\"put bottles dustbin\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n  \u003C\u002Ftable>\n\n- **Task**: put object in cabinet, put bottles in dustbin\n- **Challenge**: Long-horizon transport, partial occlusion, and constrained final placement\n- **Ours**:\n  - ✅ Maintains long-horizon scene coherence\n  - ✅ Handles occlusion without object drift\n  - ✅ Produces stable constrained placement\n\n### **Out-of-Distribution Generalization**\n\n> To test generalization beyond benchmark initial states, we use **GPT-Image-2-created initial scenes** with original robot action sequences and let **BWM** autoregressively roll out the future under object appearance shifts.\n\n  \u003Ctable align=\"center\" >\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fout_of_distribution\u002Fepisode100.gif\" alt=\"ood episode100\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fout_of_distribution\u002Fepisode100-1.gif\" alt=\"ood episode100 variant 1\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fout_of_distribution\u002Fepisode100-3.gif\" alt=\"ood episode100 variant 3\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n    \u003Ctr>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fout_of_distribution\u002Fepisode33.gif\" alt=\"ood episode33\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fout_of_distribution\u002Fepisode33-1.gif\" alt=\"ood episode33 variant 1\" width=\"260\">\u003C\u002Ftd>\n      \u003Ctd>\u003Cimg src=\"assets\u002Fout_of_distribution\u002Fepisode33-5.gif\" alt=\"ood episode33 variant 5\" width=\"260\">\u003C\u002Ftd>\n    \u003C\u002Ftr>\n  \u003C\u002Ftable>\n\n- **Task**: shake bottle, put object in cabinet\n- **Challenge**: Novel initial scenes and object appearance shifts\n- **Ours**:\n  - ✅ Generalizes to GPT-Image-2-created initial scenes\n  - ✅ Preserves action-conditioned dynamics\n  - ✅ Maintains coherent robot-object interaction\n\n---\n\n## 🛠️ Usage\n\n### Quick Start: Video Generation Inference\n\n#### Environment Setup\n\n```bash\n# Create conda environment\nconda create -n BWM python=3.10.20\nconda activate BWM\n\n# Install PyTorch with CUDA support\npip install torch==2.8.0 torchvision==0.23.0 torchaudio==2.8.0 --index-url https:\u002F\u002Fdownload.pytorch.org\u002Fwhl\u002Fcu128\n\n# Install DiffSynth-Studio\npip install diffsynth==2.0.11\n\n# Install dependencies\npip install -r requirements.txt\n```\n\n#### Model Weights\n\nDownload the [Wan2.2-TI2V-5B](https:\u002F\u002Fwww.modelscope.cn\u002Fmodels\u002FWan-AI\u002FWan2.2-TI2V-5B) base model from [ModelScope](https:\u002F\u002Fwww.modelscope.cn):\n\n```bash\nmodelscope download --model Wan-AI\u002FWan2.2-TI2V-5B --local_dir models\u002FWan2.2-TI2V-5B\n```\n\nDownload the [BWM checkpoint](https:\u002F\u002Fhuggingface.co\u002FBLM-Lab\u002FBoundless-World-Model) from [Hugging Face](https:\u002F\u002Fhuggingface.co):\n\n```bash\nhf download BLM-Lab\u002FBoundless-World-Model step-12000.safetensors --local-dir ckpt\u002FBLM\n```\n\n#### Run Inference\n\nThe demo metadata, videos, actions, and normalization statistics are already included under `demo\u002F`.\n\nSet local paths before running inference:\n\n```bash\ncp scripts\u002Flocal.example.sh scripts\u002Flocal.sh\n```\n\nUpdate `MODEL_PATHS` and `CKPT_PATH` in `scripts\u002Flocal.sh`, then run:\n\n```bash\nbash scripts\u002Finfer_example.sh\n```\n\n## 🏋️ Training\n\nComing soon !\n\n---\n\n## 🙏 Acknowledgements\n\nThis project builds upon the following open-source projects and benchmarks.\nWe thank these teams for their contributions:\n\n- Wan2.2: https:\u002F\u002Fgithub.com\u002FWan-Video\u002FWan2.2\n- DiffSynth-Studio: https:\u002F\u002Fgithub.com\u002Fmodelscope\u002FDiffSynth-Studio\n- WorldArena: https:\u002F\u002Fgithub.com\u002Ftsinghua-fib-lab\u002FWorldArena\u002F\n- ABot-PhysWorld: https:\u002F\u002Fgithub.com\u002Famap-cvlab\u002FABot-PhysWorld\n\nWe also acknowledge the following engineering contributions:\n\n- Wentao Tan: basic architecture design · [Email](mailto:tan.wt.lucky@gmail.com) · [GitHub](https:\u002F\u002Fgithub.com\u002FFutureTwT)\n- Zengrong Lin: core code implementation · [Email](mailto:zengronglin@tongji.edu.cn) · [GitHub](https:\u002F\u002Fgithub.com\u002Fzzezze)\n- Yang Sun: code refactoring and software maintainability · [Email](mailto:1006954899@qq.com) · [GitHub](https:\u002F\u002Fgithub.com\u002FDandelionWow)\n\n---\n\n\n## 📜 Citing\n\nIf you find **BWM** is useful in your research or applications, please consider giving us a **star** 🌟.\n\n---\n",2,"2026-06-11 04:11:22","high_star"]