模型
模型目录、CLI 命令、runtime 就绪状态，以及如何为一次运行选择模型。
使用本页
当你需要选择模型、确认本机能不能跑，或者理解某个支持状态是什么意思时，先看这一页。下面的 catalog 名称是模型分组，不是单独命令；日常使用优先从 CLI 开始：
| 目标 | 命令 |
|---|---|
| 查看已知模型 id 和 aliases | worldfoundry-eval zoo models --json |
| 查看单个模型 manifest | worldfoundry-eval zoo model-show --model-id <model-id> --include-manifest --json |
| 检查本地 checkpoint、凭据和资产 blocker | worldfoundry-eval zoo model-download --model-id <model-id> --check-local --json |
| 列出 runtime runner | worldfoundry-eval models runtime-runners --json |
跑推理
准备 requests.jsonl 后，用 pipeline runner 生成 artifact：
worldfoundry-eval evaluate \
--mode model \
--model-id <model-id> \
--model-runner worldfoundry:pipeline \
--model-manifest-dir worldfoundry/data/models/catalog \
--requests-path tmp/requests.jsonl \
--output-dir tmp/model_eval \
--metric artifact_count \
--required-artifact video \
--json完整推理流程（环境、checkpoint、Studio）见 推理指南。
输出文件
| 文件 | 用途 |
|---|---|
run_manifest.json | run id、model、benchmark metadata、时间戳、输出路径 |
results.jsonl | 每样本状态、artifact、耗时 |
metrics/summary.json | 聚合指标 |
scorecard.json | leaderboard eligibility 与 blocking reason |
跑失败时
zoo model-show --include-manifest— integration status 与 blockerzoo model-download --check-local— checkpoint、credential 和 asset readiness- 对照 benchmark 页面确认 dataset、API key 和生成结果目录要求
- 对照下方 model registry，确认该 model id 是否已有 runner 证据
选择模型
投入 GPU 时间前，按顺序确认这些问题：
| 问题 | 检查方式 |
|---|---|
| WorldFoundry 是否认识这个模型？ | 它出现在 worldfoundry-eval zoo models --json 中，并且有稳定的 model_id。 |
| 这个 repo 能否路由到该模型？ | Manifest 能解析到 runner 和 pipeline binding。runtime_ported 表示 route 已存在，但真实 checkpoint/API/simulator parity 可能还没完成。 |
| 你的环境能否运行它？ | zoo model-download --check-local、models assets 和 benchmark 页面会说明需要的 checkpoints、API keys、datasets 或 simulators。 |
| 它是否对你的任务 benchmark-ready？ | 需要看该模型是否有对应 benchmark 或 artifact 类型的证据。视频、3D、world-action 和 VLA 任务需要不同证据。 |
怎么读 catalog 行
| 列 | 含义 |
|---|---|
| Family | 公开项目或 provider 家族；不同 variant 仍可能有独立 manifest。 |
| Capability | 模型在 WorldFoundry 语义下能生成或预测什么。 |
| Runtime | local、official repo、subprocess、hosted API 或混合形态。 |
| Evidence target | 升级公开声明前最有用的下一步证据。 |
当前覆盖摘要
WorldFoundry 按 manifest、runner 和 benchmark evidence 三层跟踪模型支持状态。先用这个摘要找到合适的模型分组，再用 zoo model-show 查看具体 model id。
| Catalog directory | 条目数 | 覆盖内容 | 代表 id |
|---|---|---|---|
worldfoundry/data/models/catalog/video/ | 48 | 视频、音频与多模态生成模型 catalog（原 video_open 与 video_pipeline_runner 合并）。含开源 provenance 记录与带 pipeline binding 的 runtime 集成。 | wan2.1、hunyuanvideo、animatediff、zeroscope、step-video-t2v、sana、vmem、krea-realtime-video。 |
worldfoundry/data/models/catalog/world_models/ | 69 | 交互式世界模型、simulator、导航与 camera-control 模型 catalog（原 world_models 与 world_pipeline_runner 合并）。 | matrix-game-2、lingbot-world、hunyuanworld-voyager、diamond、dino-wm、leworldmodel、starwm、irasim。 |
worldfoundry/data/models/catalog/three_d_four_d/ | 47 | 3D/4D generation、reconstruction、point-cloud、geometry、scene-representation 和 perception-prior 记录。状态分布：25 integrated、10 runtime_ported、10 blocked、2 planned；多数通过 pipeline binding 解析。 | flashworld、fantasyworld、dust3r、monst3r、splatt3r、pixelsplat、vggt-omega、dvlt。 |
worldfoundry/data/models/catalog/vla_va_wam/ | 42 | VLA、VA、WAM、robot-policy 和 embodied-action 记录。所有条目都有 runner target 和 binding；多数是 runtime_ported，完整 policy/runtime 声明仍受 checkpoint 与 simulator 验证约束。 | openvla、openpi、pi0-fast、smolvla、xvla、eo1、tdmpc、vqbet、wall-oss、act。 |
模型来源索引
这个索引结合 worldfoundry/data/models/catalog 的 manifest 与本轮确认到的官方论文、项目页和源码链接。如果没有单独发布 paper、project page 或 code repository，则使用最接近的官方 model card、source page 或 project URL。
开源视频模型
上游视频生成模型族，以及仓内 runtime / metadata 记录。支持层级不一：有的已有本地生成验证路径，有的是 checkpoint-gated wrapper、hosted/API 形态，或仅 provenance。跑之前先看 catalog readiness 和 zoo model-show。
Wan2.1 (wan2.1)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2503.20314
- 项目主页: https://wan.video/
- GitHub: https://github.com/Wan-Video/Wan2.1
Wan2.2 (wan2.2)
- 能力: text-to-video, image-to-video, text-image-to-video
- 论文: https://arxiv.org/abs/2503.20314
- 项目主页: https://wan.video/
- GitHub: https://github.com/Wan-Video/Wan2.2
HunyuanVideo (hunyuanvideo)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2412.03603
- 项目主页: https://aivideo.hunyuan.tencent.com/
- GitHub: https://github.com/Tencent-Hunyuan/HunyuanVideo
LTX-Video (ltx-video)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2501.00103
- 项目主页: https://ltx.video/
- GitHub: https://github.com/Lightricks/LTX-Video
LTX-2.x (ltx-2.x)
- 能力: text-to-video, image-to-video, video-to-video, audio-video-generation
- 论文: https://arxiv.org/abs/2601.03233
- 项目主页: https://ltx.video/
- GitHub: https://github.com/Lightricks/LTX-Video
CogVideoX (cogvideox)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2408.06072
- 项目主页: https://yzy-thu.github.io/CogVideoX-demo/
- GitHub: https://github.com/THUDM/CogVideo
Open-Sora (open-sora)
- 能力: text-to-video
- 论文: https://arxiv.org/abs/2412.20404
- 项目主页: https://hpcaitech.github.io/Open-Sora/
- GitHub: https://github.com/hpcaitech/Open-Sora
Open-Sora-Plan (open-sora-plan)
- 能力: text-to-video
- 论文: https://arxiv.org/abs/2412.00131
- 项目主页: https://github.com/PKU-YuanGroup/Open-Sora-Plan
- GitHub: https://github.com/PKU-YuanGroup/Open-Sora-Plan
Mochi-1 (mochi-1)
- 能力: text-to-video
- 论文: https://huggingface.co/genmo/mochi-1-preview
- 项目主页: https://www.genmo.ai/play
- GitHub: https://github.com/genmoai/mochi
Allegro (allegro)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2410.15458
- 项目主页: https://rhymes.ai/allegro_gallery
- GitHub: https://github.com/rhymes-ai/Allegro
EasyAnimate (easyanimate)
- 能力: text-to-video, image-to-video, video-to-video
- 论文: https://arxiv.org/abs/2405.18991
- 项目主页: https://easyanimate.github.io/
- GitHub: https://github.com/aigc-apps/EasyAnimate
VideoCrafter (videocrafter)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2401.09047
- 项目主页: https://ailab-cvc.github.io/videocrafter2/
- GitHub: https://github.com/VideoCrafter/VideoCrafter
DynamiCrafter (dynamicrafter)
- 能力: image-to-video
- 论文: https://arxiv.org/abs/2310.12190
- 项目主页: https://doubiiu.github.io/projects/DynamiCrafter/
- GitHub: https://github.com/Doubiiu/DynamiCrafter
I2VGen-XL (i2vgen-xl)
- 能力: image-to-video, text-to-video
- 论文: https://arxiv.org/abs/2311.04145
- 项目主页: https://i2vgen-xl.github.io/
- GitHub: https://github.com/ali-vilab/i2vgen-xl
ModelScopeT2V (modelscope-t2v)
- 能力: text-to-video
- 论文: https://arxiv.org/abs/2308.06571
- 项目主页: https://modelscope.cn/models/damo/text-to-video-synthesis/summary
- GitHub: https://github.com/modelscope/modelscope
HunyuanVideo 1.5 (hunyuanvideo-1.5)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2511.18870
- 项目主页: https://hunyuan.tencent.com/video/zh?tabIndex=0
- GitHub: https://github.com/Tencent-Hunyuan/HunyuanVideo-1.5
SkyReels-V2 (skyreels-v2)
- 能力: text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2504.13074
- 项目主页: https://github.com/SkyworkAI/SkyReels-V2
- GitHub: https://github.com/SkyworkAI/SkyReels-V2
MAGI-1 (magi-1)
- 能力: image-to-video, video-generation
- 论文: https://arxiv.org/abs/2505.13211
- 项目主页: https://magi.sand.ai/
- GitHub: https://github.com/SandAI-org/MAGI-1
FramePack (framepack)
- 能力: image-to-video, long-video
- 论文: https://arxiv.org/abs/2504.12626
- 项目主页: https://lllyasviel.github.io/frame_pack_gitpage/
- GitHub: https://github.com/lllyasviel/FramePack
Step-Video-T2V (step-video-t2v)
- 能力: text-to-video, video-generation
- 论文: https://arxiv.org/abs/2502.10248
- 项目主页: https://yuewen.cn/videos
- GitHub: https://github.com/stepfun-ai/Step-Video-T2V
Open-MAGVIT2 (open-magvit2)
- 能力: video-tokenization, autoregressive-generation
- 论文: https://arxiv.org/abs/2409.04410
- 项目主页: https://huggingface.co/TencentARC/Open-MAGVIT2
- GitHub: https://github.com/TencentARC/Open-MAGVIT2
Show-o (show-o)
- 能力: multimodal-generation, text-to-image, video-generation
- 论文: https://arxiv.org/abs/2408.12528
- 项目主页: https://showlab.github.io/Show-o/
- GitHub: https://github.com/showlab/Show-o
AnimateDiff (animatediff)
- 能力: text-to-video, classic-unet-diffusion
- 论文: https://arxiv.org/abs/2307.04725
- 项目主页: https://animatediff.github.io/
- GitHub: https://github.com/guoyww/AnimateDiff
ZeroScope (zeroscope)
- 能力: text-to-video, classic-unet-diffusion
- 论文: https://huggingface.co/cerspense/zeroscope_v2_576w
- 项目主页: https://huggingface.co/cerspense/zeroscope_v2_576w
- GitHub: https://huggingface.co/cerspense/zeroscope_v2_576w
世界模型
交互式、相机控制、导航、驾驶与长时程仿真系统，也包括 provenance / metadata-only 条目。能力标签是契约提示：action-controlled 模型应保留 action traces，camera-controlled 模型应暴露 camera metadata，长时程系统需要能保持序列顺序的 artifact 命名。
Cosmos Predict 2.5 (cosmos-predict-2.5)
- 能力: text-to-video, image-to-video, video-to-video, world-generation
- 论文: https://arxiv.org/abs/2511.00062
- 项目主页: https://research.nvidia.com/labs/cosmos-lab/cosmos-predict2.5/
- GitHub: https://github.com/nvidia-cosmos/cosmos-predict2.5
Cosmos Transfer 2.5 (cosmos-transfer-2.5)
- 能力: controlled-video, world-generation, spatial-control
- 论文: https://arxiv.org/abs/2511.00062
- 项目主页: https://research.nvidia.com/labs/cosmos-lab/cosmos-transfer2.5/
- GitHub: https://github.com/nvidia-cosmos/cosmos-transfer2.5
DreamDojo (dreamdojo)
- 能力: interactive-world-model, robot-world-model, action-conditioned-video2world, long-horizon
- 论文: https://arxiv.org/abs/2602.06949
- 项目主页: https://dreamdojo-world.github.io/
- GitHub: https://github.com/NVIDIA/DreamDojo
Matrix-Game 1.0 (matrix-game-1)
- 能力: interactive-world-model, game-video, action-control
- 论文: https://arxiv.org/abs/2506.18701
- 项目主页: https://matrix-game-homepage.github.io/
- GitHub: https://github.com/SkyworkAI/Matrix-Game
Matrix-Game 2.0 (matrix-game-2)
- 能力: interactive-world-model, streaming, game-video, action-control
- 论文: https://arxiv.org/abs/2508.13009
- 项目主页: https://matrix-game-v2.github.io/
- GitHub: https://github.com/SkyworkAI/Matrix-Game
Matrix-Game 3.0 (matrix-game-3)
- 能力: interactive-world-model, streaming, long-horizon, game-video
- 论文: https://arxiv.org/abs/2604.08995
- 项目主页: https://matrix-game-v3.github.io/
- GitHub: https://github.com/SkyworkAI/Matrix-Game
LingBot-World (lingbot-world)
- 能力: interactive-world-model, camera-control, world-video
- 论文: https://arxiv.org/abs/2601.20540
- 项目主页: https://technology.robbyant.com/lingbot-world
- GitHub: https://github.com/robbyant/lingbot-world
HunyuanWorld 1.0 (hunyuanworld-1)
- 能力: text-to-3d-world, image-to-3d-world, panorama, mesh-export
- 论文: https://arxiv.org/abs/2507.21809
- 项目主页: https://3d-models.hunyuan.tencent.com/world/
- GitHub: https://github.com/Tencent-Hunyuan/HunyuanWorld-1.0
HunyuanWorld-Voyager (hunyuanworld-voyager)
- 能力: image-to-video, 3d-reconstruction, world-generation, long-range
- 论文: https://arxiv.org/abs/2506.04225
- 项目主页: https://3d-models.hunyuan.tencent.com/voyager/
- GitHub: https://github.com/Tencent-Hunyuan/HunyuanWorld-Voyager
HY-WorldPlay (hy-worldplay)
- 能力: interactive-world-model, streaming, real-time, camera-action-control
- 论文: https://arxiv.org/abs/2512.14614
- 项目主页: https://3d-models.hunyuan.tencent.com/world/
- GitHub: https://github.com/Tencent-Hunyuan/HY-WorldPlay
HunyuanWorld-Mirror (hunyuanworld-mirror)
- 能力: 3d-reconstruction, world-mirror, image-to-3d, video-to-3d
- 论文: https://3d-models.hunyuan.tencent.com/world/worldMirror1_0/HYWorld_Mirror_Tech_Report.pdf
- 项目主页: https://3d-models.hunyuan.tencent.com/world/
- GitHub: https://github.com/Tencent-Hunyuan/HunyuanWorld-Mirror
HY-World 2.0 (hy-world-2.0)
- 能力: text-to-3d-world, image-to-3d-world, video-to-3d-world, simulation
- 论文: https://arxiv.org/abs/2604.14268
- 项目主页: https://3d-models.hunyuan.tencent.com/world/
- GitHub: https://github.com/Tencent-Hunyuan/HY-World-2.0
InSpatio-WorldFM (worldfm)
- 能力: 3d-world-model, spatial-intelligence, frame-generation
- 论文: https://arxiv.org/abs/2603.11911
- 项目主页: https://inspatio.github.io/worldfm/
- GitHub: https://github.com/inspatio/worldfm
Infinite-World (infinite-world)
- 能力: interactive-world-model, long-horizon, image-to-video, memory
- 论文: https://arxiv.org/abs/2602.02393
- 项目主页: https://huggingface.co/MeiGen-AI/Infinite-World
- GitHub: https://github.com/MeiGen-AI/Infinite-World
WorldCam (worldcam)
- 能力: interactive-world-model, camera-pose-control, game-video, long-horizon
- 论文: https://arxiv.org/abs/2603.16871
- 项目主页: https://cvlab-kaist.github.io/WorldCam/
- GitHub: https://github.com/cvlab-kaist/WorldCam
CameraCtrl (cameractrl)
- 能力: camera-control, text-to-video, video-diffusion-plugin
- 论文: https://arxiv.org/abs/2404.02101
- 项目主页: https://hehao13.github.io/projects-CameraCtrl/
- GitHub: https://github.com/hehao13/CameraCtrl
MotionCtrl (motionctrl)
- 能力: camera-control, object-motion-control, text-to-video, image-to-video
- 论文: https://arxiv.org/abs/2312.03641
- 项目主页: https://wzhouxiff.github.io/projects/MotionCtrl/
- GitHub: https://github.com/TencentARC/MotionCtrl
ViewCrafter (viewcrafter)
- 能力: novel-view-synthesis, camera-control, sparse-view, 3d-reconstruction
- 论文: https://arxiv.org/abs/2409.02048
- 项目主页: https://drexubery.github.io/ViewCrafter/
- GitHub: https://github.com/Drexubery/ViewCrafter
IRASim (irasim)
- 能力: robotics-world-model, video-generation, action-conditioned-generation
- 论文: https://arxiv.org/abs/2406.14540
- 项目主页: https://gen-irasim.github.io/
- GitHub: https://github.com/bytedance/IRASim
Pandora (pandora)
- 能力: diffusion-world-model, world-generation, text-to-video
- 论文: https://arxiv.org/abs/2406.09455
- 项目主页: https://world-model.maitrix.org/
- GitHub: https://github.com/maitrix-org/Pandora
3D 和 4D 模型
3D/4D 生成、重建、点云、几何与 perception-prior 模型，也包括 provenance / metadata-only 条目。需要可度量的 3D 输出时，优先选带显式 geometry、camera 或 asset metadata 的条目；部分行在 WorldFoundry 可跑之前仍需 port plan。
MVDiffusion (mvdiffusion)
- 能力: correspondence-aware multiview image generation and scene texturing, 3d
- 论文: https://arxiv.org/abs/2307.01097
- 项目主页: https://mvdiffusion.github.io/
- GitHub: https://github.com/Tangshitao/MVDiffusion
4D-GS (4d-gs)
- 能力: real-time dynamic scene rendering, 4d
- 论文: https://arxiv.org/abs/2310.08528
- 项目主页: https://guanjunwu.github.io/4dgs/
- GitHub: https://github.com/hustvl/4DGaussians
MonST3R (monst3r)
- 能力: geometry estimation in the presence of motion, 4d
- 论文: https://monst3r-project.github.io/files/monst3r_paper.pdf
- 项目主页: https://monst3r-project.github.io/
- GitHub: https://github.com/Junyi42/monst3r
Shape of Motion (shape-of-motion)
- 能力: single-video 4D reconstruction, 4d
- 论文: https://arxiv.org/abs/2407.13764
- 项目主页: https://shape-of-motion.github.io/
- GitHub: https://github.com/vye16/shape-of-motion
FlashWorld (flashworld)
- 能力: fast 3D scene generation from image or text, 3d
- 论文: https://arxiv.org/abs/2510.13678
- 项目主页: https://imlixinyang.github.io/FlashWorld-Project-Page/
- GitHub: https://github.com/imlixinyang/FlashWorld
FantasyWorld (fantasyworld)
- 能力: geometry-consistent world modeling, 3d
- 论文: https://arxiv.org/abs/2509.21657
- 项目主页: https://fantasy-amap.github.io/fantasy-world/
- GitHub: https://github.com/Fantasy-AMAP/fantasy-world
WonderWorld (wonderworld)
- 能力: interactive 3D scene generation from a single image, 3d
- 论文: https://arxiv.org/abs/2406.09394
- 项目主页: https://kovenyu.com/wonderworld/
- GitHub: https://github.com/KovenYu/WonderWorld
NeoVerse (neoverse)
- 能力: 4D world modeling from in-the-wild monocular videos, 4d
- 论文: https://arxiv.org/abs/2601.00393
- 项目主页: https://neoverse-4d.github.io/
- GitHub: https://github.com/IamCreateAI/NeoVerse
ReCamMaster (recammaster)
- 能力: camera-controlled video rerendering, 4d
- 论文: https://arxiv.org/abs/2503.11647
- 项目主页: https://jianhongbai.github.io/ReCamMaster/
- GitHub: https://github.com/KlingAIResearch/ReCamMaster
Lyra (lyra)
- 能力: open generative 3D and 4D world models, 3d_4d
- 论文: https://openreview.net/forum?id=tIVCfVnIHo
- 项目主页: https://research.nvidia.com/labs/sil/projects/lyra2/
- GitHub: https://github.com/nv-tlabs/lyra
Stable Virtual Camera (stable-virtual-camera)
- 能力: generative novel view synthesis with camera control, 3d
- 论文: https://arxiv.org/abs/2503.14489
- 项目主页: https://stable-virtual-camera.github.io/
- GitHub: https://github.com/Stability-AI/stable-virtual-camera
Splatt3R (splatt3r)
- 能力: 3d-reconstruction, gaussian-splatting
- 论文: https://arxiv.org/abs/2408.13912
- 项目主页: https://splatt3r.active.vision/
- GitHub: https://github.com/btsmart/splatt3r
pixelSplat (pixelsplat)
- 能力: 3d-reconstruction, gaussian-splatting, novel-view-synthesis
- 论文: https://arxiv.org/abs/2312.12337
- 项目主页: https://dcharatan.github.io/pixelsplat/
- GitHub: https://github.com/dcharatan/pixelsplat
VLA、VA 和 WAM 模型
这类模型评测的是行为，而不只是图像质量。一次可用 run 需要 observation、语言或任务指令、action output、simulator/dataset context，以及能映射到归一化指标的 success signal。
openvla 是这一组 checkpoint-verified embodied-action baseline。MolmoAct2 也有仓内 WorldFoundry pipeline/runtime surface，用于 DROID 和 bimanual YAM 的 action-trace contract；checkpoint-backed GPU parity 仍待验证。原来的 generic LeRobot policy-zoo surface 已取消；lerobot/src/lerobot/policies 里的 policy 现在按具体算法追踪各自上游来源，例如 ACT、Diffusion Policy、TD-MPC、VQ-BeT、pi0/pi0.5/pi0-FAST、SmolVLA、GR00T、X-VLA、WALL-OSS、EO-1、MolmoAct2、Multi-task DiT、Gaussian Actor 和 RTC。多数 VLA/VA/WAM 行当前是 runtime_ported：它们已有 WorldFoundry route contract，但 task support 仍受 checkpoint、simulator 和真实推理证据约束。
OpenVLA (openvla)
- 能力: vla, vla.action_prediction, vla.policy_rollout
- 论文: https://arxiv.org/abs/2406.09246
- 项目主页: https://openvla.github.io/
- GitHub: https://github.com/openvla/openvla
OpenPI (openpi)
- 能力: vla, vla.policy_rollout, robot_policy
- 论文: https://arxiv.org/abs/2410.24164
- 项目主页: https://www.physicalintelligence.company/blog/pi0
- GitHub: https://github.com/Physical-Intelligence/openpi
MolmoAct2 (molmoact2)
- 能力: vla, vla.action_prediction, vla.policy_rollout, robot_policy
- 论文: https://arxiv.org/abs/2605.02881
- 项目主页: https://allenai.org/blog/molmoact2
- GitHub: https://github.com/allenai/molmoact2
TD-MPC (tdmpc)
- 能力: model_predictive_control, robot_policy, policy_rollout
- 论文: https://www.nicklashansen.com/td-mpc/
- 项目主页: https://www.nicklashansen.com/td-mpc/
- GitHub: https://github.com/nicklashansen/tdmpc
VQ-BeT (vqbet)
- 能力: behavior_generation, imitation_learning, robot_policy
- 论文: https://arxiv.org/abs/2403.03181
- 项目主页: https://sjlee.cc/vq-bet/
- GitHub: https://github.com/jayLEE0301/vq_bet_official
SmolVLA (smolvla)
- 能力: vla, vla.policy_rollout, robot_policy
- 论文: https://arxiv.org/abs/2506.01844
- 项目主页: https://huggingface.co/lerobot/smolvla_base
- GitHub: https://github.com/huggingface/lerobot
X-VLA (xvla)
- 能力: vla, cross_embodiment_vla, vla.policy_rollout
- 论文: https://huggingface.co/papers/2510.10274
- 项目主页: https://huggingface.co/collections/2toINF/x-vla
- GitHub: https://github.com/2toinf/X-VLA
EO-1 (eo1)
- 能力: vla, embodied_vla, vla.policy_rollout
- 论文: https://arxiv.org/abs/2508.21112
- 项目主页: https://eo-robotics.ai/eo-1
- GitHub: https://github.com/EO-Robotics/EO
Being-H0.5 (being-h05)
- 能力: vla, vla.policy_rollout, robot_policy, cross_embodiment_policy
- 论文: https://arxiv.org/abs/2601.12993
- 项目主页: https://research.beingbeyond.com/being-h05
- GitHub: https://github.com/BeingBeyond/Being-H
NVIDIA DreamZero (dreamzero)
- 能力: wam, world_action_model, wam.world_action_modeling, robot_policy, video_prediction
- 论文: https://arxiv.org/abs/2602.15922
- 项目主页: https://dreamzero0.github.io/
- GitHub: https://github.com/dreamzero0/dreamzero
NVIDIA GR00T (gr00t)
- 能力: vla, vla.policy_rollout, humanoid_policy
- 论文: https://arxiv.org/abs/2503.14734
- 项目主页: https://developer.nvidia.com/isaac/gr00t
- GitHub: https://github.com/NVIDIA/Isaac-GR00T
StarVLA (starvla)
- 能力: vla, wam, vla.policy_rollout, wam.world_action_modeling
- 论文: https://arxiv.org/abs/2604.05014
- 项目主页: https://starvla.github.io/
- GitHub: https://github.com/starVLA/starVLA
LingBot-VA (lingbot-va)
- 能力: vla, va, vam, wam, vla.policy_rollout, vam.video_action_modeling, wam.world_action_modeling, robot_policy
- 论文: https://arxiv.org/abs/2601.21998
- 项目主页: https://technology.robbyant.com/lingbot-va
- GitHub: https://github.com/Robbyant/lingbot-va
Latent Action Pretraining (lapa)
- 能力: va, vam, latent_action, video_action_modeling
- 论文: https://arxiv.org/abs/2410.11758
- 项目主页: https://latentactionpretraining.github.io/
- GitHub: https://github.com/LatentActionPretraining/LAPA
Octo (octo)
- 能力: vla, vla.policy_rollout, robot_policy, generalist_policy
- 论文: https://octo-models.github.io/paper.pdf
- 项目主页: https://octo-models.github.io/
- GitHub: https://github.com/octo-models/octo
RT-1 (rt-1)
- 能力: vla, vla.policy_rollout, robot_policy, robotics_transformer
- 论文: https://arxiv.org/abs/2212.06817
- 项目主页: https://robotics-transformer1.github.io/
- GitHub: https://github.com/google-research/robotics_transformer
Diffusion Policy (diffusion-policy)
- 能力: visuomotor_policy, robot_policy, action_diffusion, policy_rollout
- 论文: https://diffusion-policy.cs.columbia.edu/#paper
- 项目主页: https://diffusion-policy.cs.columbia.edu/
- GitHub: https://github.com/real-stanford/diffusion_policy
Action Chunking Transformer (act)
- 能力: action_chunking_policy, robot_policy, imitation_policy, policy_rollout
- 论文: https://arxiv.org/abs/2304.13705
- 项目主页: https://tonyzhaozh.github.io/aloha/
- GitHub: https://github.com/tonyzhaozh/act
RoboFlamingo (roboflamingo)
- 能力: vla, vla.policy_rollout, vlm_policy, robot_policy
- 论文: https://arxiv.org/abs/2311.01378
- 项目主页: https://github.com/RoboFlamingo/RoboFlamingo
- GitHub: https://github.com/RoboFlamingo/RoboFlamingo