{"name":"models","generated_at":"2026-09-15T12:59:00.702Z","schema_version":"0.1","source":"https://robo.sinanlab.com","license":"CC BY 4.0 · 注明\"数据：司南实验室 robo.sinanlab.com\"","note":"未核实的字段为 null；每条记录的 evidence 数组给出字段来源。只陈述测量与来源，不含推荐。","count":40,"items":[{"id":"gr00t-n1.7","name":"Isaac GR00T N1.7","org":"NVIDIA","release_date":"2026-04","params_b":3,"arch_type":"dual_system","arch_notes":"Action Cascade:System 2 为 VLM(处理图像token与语言指令,输出高层action token);System 1 为 32 层 Diffusion Transformer,结合实时机器人状态生成运动指令","license":"NVIDIA Open Model License（权重）；Apache-2.0（Isaac-GR00T 代码）","commercial_ok":true,"weights_url":"https://huggingface.co/nvidia/GR00T-N1.7-3B","code_url":"https://github.com/NVIDIA/Isaac-GR00T","paper_url":null,"modalities_in":["rgb","language","state"],"action_space":"关节/全身(人形为主)","cross_embodiment":"以人形为主(待核实其他构型支持)","pretrain_hours":null,"target_embodiments":["humanoid"],"evidence":[{"field":"license,commercial_ok","url":"https://huggingface.co/nvidia/GR00T-N1.7-3B","source_type":"hub","fetched":"2026-09-03","note":"NVIDIA Open Model License Agreement; ready for commercial/non-commercial use"},{"field":"license","url":"https://raw.githubusercontent.com/NVIDIA/Isaac-GR00T/main/LICENSE","source_type":"official","fetched":"2026-09-03","note":"代码 Apache License 2.0"},{"field":"params_b,weights_url","url":"https://huggingface.co/api/models/nvidia/GR00T-N1.7-3B","source_type":"hub","fetched":"2026-09-03","note":"safetensors 3,144,016,000 参数"},{"field":"release_date","url":"https://forums.developer.nvidia.com/t/early-access-isaac-gr00t-n1-7-open-reasoning-vla-model-for-humanoid-robotics/366916","source_type":"official","fetched":"2026-09-03","note":"Early Access 公告 2026-04-17"},{"field":"release_date,params_b","url":"https://huggingface.co/blog/nvidia/gr00t-n1-7","source_type":"hub","fetched":"2026-09-03","note":"April 17, 2026; 3B open reasoning VLA"},{"field":"code_url,weights_url","url":"https://raw.githubusercontent.com/NVIDIA/Isaac-GR00T/main/README.md","source_type":"official","fetched":"2026-09-03","note":"GA 已发布；nvidia/GR00T-N1.7-3B"}]},{"id":"gr00t-n1.6-3b","name":"GR00T N1.6 3B","org":"NVIDIA","release_date":"2026 (待核实)","params_b":3.29,"arch_type":"dual_system","arch_notes":"N1 系列前代版本;Eagle-2 类 VLM 做慢思考 + DiT 动作专家做快反应","license":"模型卡标 NVIDIA License（链接 LICENSE 文件，条款待逐字核实）","commercial_ok":null,"weights_url":"https://huggingface.co/nvidia/GR00T-N1.6-3B","code_url":null,"paper_url":null,"modalities_in":["rgb","language","state"],"action_space":"关节/全身","cross_embodiment":"待核实","pretrain_hours":null,"target_embodiments":["humanoid"],"evidence":[{"field":"weights_url","url":"https://huggingface.co/nvidia/GR00T-N1.6-3B","source_type":"hub","fetched":"2026-08-31"},{"field":"params_b","url":"https://huggingface.co/api/models/nvidia/GR00T-N1.6-3B","source_type":"hub","fetched":"2026-09-05","note":"safetensors 3,286,608,832；createdAt 2025-12-01；模型卡 API 无 license 字段"},{"field":"license","url":"https://huggingface.co/nvidia/GR00T-N1.6-3B","source_type":"hub","fetched":"2026-09-05","note":"模型卡页面写 Nvidia License，具体条款需读 LICENSE 文件"}]},{"id":"lingbot-vla-2.0","name":"LingBot-VLA 2.0","org":"蚂蚁灵波科技 (Ant Group)","release_date":"2026-07","params_b":6,"arch_type":"待核实","arch_notes":"预训练融入 6 万小时真实物理数据;支持头部、腰部、末端执行器与移动底盘等自由度","license":"Apache-2.0","commercial_ok":true,"weights_url":"https://huggingface.co/robbyant/lingbot-vla-v2-6b","code_url":"https://github.com/robbyant/lingbot-vla-v2","paper_url":null,"modalities_in":["rgb","language","state"],"action_space":"多构型(单臂/双臂、双足/轮式)","cross_embodiment":"是：预训练覆盖 20 种机器人构型（官方）；媒体称 17 家厂商","pretrain_hours":60000,"target_embodiments":["humanoid","dual_arm","single_arm","wheeled"],"evidence":[{"field":"license,params_b,weights_url,code_url,release_date","url":"https://github.com/robbyant/lingbot-vla-v2","source_type":"official","fetched":"2026-09-03","note":"Apache-2.0; 6B; 2026-07-08 权重发布"},{"field":"license,pretrain_hours,cross_embodiment","url":"https://huggingface.co/robbyant/lingbot-vla-v2-6b/raw/main/README.md","source_type":"hub","fetched":"2026-09-03","note":"Apache-2.0; 60,000 hours; 20 robot configurations"},{"field":"params_b,release_date","url":"https://huggingface.co/api/models/robbyant/lingbot-vla-v2-6b","source_type":"hub","fetched":"2026-09-03","note":"safetensors ≈6.38B F32; created 2026-07-07"},{"field":"pretrain_hours,cross_embodiment","url":"https://arxiv.org/abs/2607.06403","source_type":"literature","fetched":"2026-09-03","note":"50k h 真机 + 10k h 第一视角视频; 20 constructions"},{"field":"cross_embodiment","url":"https://www.qbitai.com/2026/07/445668.html","source_type":"news","fetched":"2026-09-03","note":"17 个品牌 20 种构型"},{"field":"cross_embodiment","url":"https://technology.robbyant.com/lingbot-vla-v2","source_type":"official","fetched":"2026-09-03","note":"官网列出 20 个具体机型"}]},{"id":"go-1","name":"GO-1 (Genie Operator-1)","org":"智元机器人 (AgiBot)","release_date":"2025-09","params_b":2.82,"arch_type":"villa","arch_notes":"Vision-Language-Latent-Action (ViLLA) 架构;向全球开发者免费开放","license":"CC BY-NC-SA 4.0","commercial_ok":false,"weights_url":"https://huggingface.co/agibot-world/GO-1","code_url":"https://github.com/OpenDriveLab/AgiBot-World","paper_url":"https://arxiv.org/abs/2503.06669","modalities_in":["rgb","language","state"],"action_space":"待核实","cross_embodiment":"支持其他本体的数据采集、微调与部署(官方表述)","pretrain_hours":null,"target_embodiments":["humanoid","dual_arm"],"evidence":[{"field":"license,commercial_ok","url":"https://huggingface.co/agibot-world/GO-1/raw/main/README.md","source_type":"hub","fetched":"2026-09-03","note":"License: CC BY-NC-SA 4.0"},{"field":"license,release_date,code_url","url":"https://raw.githubusercontent.com/OpenDriveLab/AgiBot-World/main/README.md","source_type":"official","fetched":"2026-09-03","note":"2025/09/19 GO-1 open-sourced; CC BY-NC-SA 4.0"},{"field":"params_b,release_date,weights_url","url":"https://huggingface.co/api/models/agibot-world/GO-1","source_type":"hub","fetched":"2026-09-03","note":"safetensors 2,821,541,905 BF16; created 2025-09-04"},{"field":"paper_url","url":"https://huggingface.co/agibot-world/GO-1","source_type":"hub","fetched":"2026-09-03","note":"arXiv 2503.06669"}]},{"id":"pi0","name":"π0 (pi-zero)","org":"Physical Intelligence","release_date":"2025-02","params_b":3.3,"arch_type":"vlm_plus_action_expert","arch_notes":"基于 PaliGemma 类 VLM + flow-matching 动作专家(参数量为常见公开说法,待核实)","license":"Apache-2.0（openpi 代码）；权重无独立许可证声明，仓库附 Gemma 使用条款","commercial_ok":null,"weights_url":"https://huggingface.co/lerobot/pi0_base","code_url":"https://github.com/Physical-Intelligence/openpi","paper_url":"https://arxiv.org/abs/2410.24164","modalities_in":["rgb","language","state"],"action_space":"关节(机械臂/双臂为主)","cross_embodiment":"多本体预训练(待核实清单)","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm","mobile_manipulator"],"evidence":[{"field":"license","url":"https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/LICENSE","source_type":"official","fetched":"2026-09-03","note":"代码 Apache-2.0"},{"field":"license","url":"https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/LICENSE_GEMMA.txt","source_type":"official","fetched":"2026-09-03","note":"Gemma Terms of Use 约束模型衍生物"},{"field":"weights_url,code_url","url":"https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/README.md","source_type":"official","fetched":"2026-09-03","note":"官方权重 gs://openpi-assets/checkpoints/pi0_base；README 无权重许可证章节"},{"field":"release_date","url":"https://www.pi.website/blog/openpi","source_type":"official","fetched":"2026-09-03","note":"Open Sourcing π0, February 4, 2025"},{"field":"params_b,paper_url","url":"https://arxiv.org/html/2410.24164v1","source_type":"literature","fetched":"2026-09-03","note":"total 3.3 billion parameters"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/lerobot/pi0_base","source_type":"hub","fetched":"2026-09-03","note":"LeRobot 移植版 license 标 gemma"}]},{"id":"pi0.5","name":"π0.5","org":"Physical Intelligence","release_date":"2025-09","params_b":null,"arch_type":"vlm_plus_action_expert","arch_notes":"π0 后继,强调开放世界泛化(待核实)","license":"Apache-2.0（openpi 代码）；权重无独立许可证声明，仓库附 Gemma 使用条款","commercial_ok":null,"weights_url":"https://huggingface.co/lerobot/pi05_base","code_url":"https://github.com/Physical-Intelligence/openpi","paper_url":"https://arxiv.org/abs/2504.16054","modalities_in":["rgb","language","state"],"action_space":"关节","cross_embodiment":"待核实","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm","mobile_manipulator"],"evidence":[{"field":"release_date,weights_url,code_url","url":"https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/README.md","source_type":"official","fetched":"2026-09-03","note":"Sept 2025 released pi05; gs://openpi-assets/checkpoints/pi05_base"},{"field":"paper_url","url":"https://arxiv.org/abs/2504.16054","source_type":"literature","fetched":"2026-09-03","note":"2025-04-22 提交；全文未给参数量"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/lerobot/pi05_base","source_type":"hub","fetched":"2026-09-03","note":"LeRobot 移植版 license gemma; ≈3.6B F32"},{"field":"license","url":"https://raw.githubusercontent.com/Physical-Intelligence/openpi/main/LICENSE","source_type":"official","fetched":"2026-09-03","note":"代码 Apache-2.0"}]},{"id":"smolvla","name":"SmolVLA","org":"Hugging Face (LeRobot)","release_date":"2025-06","params_b":0.45,"arch_type":"vlm_plus_action_expert","arch_notes":"紧凑型:视觉-语言模型 + 输出控制命令的动作专家;可在 MacBook 级设备运行(参数量为常见公开说法,待核实)","license":"权重模型卡未标注许可证（LeRobot 代码仓库 Apache-2.0）","commercial_ok":null,"weights_url":"https://huggingface.co/lerobot/smolvla_base","code_url":"https://github.com/huggingface/lerobot","paper_url":"https://arxiv.org/abs/2506.01844","modalities_in":["rgb","language","state"],"action_space":"关节(低成本机械臂为主)","cross_embodiment":"以 SO-100/101 等社区本体为主","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"params_b,release_date,weights_url","url":"https://huggingface.co/api/models/lerobot/smolvla_base","source_type":"hub","fetched":"2026-09-03","note":"safetensors 450,046,176; created 2025-06-01; 无 license 标签"},{"field":"params_b","url":"https://arxiv.org/html/2506.01844v1","source_type":"literature","fetched":"2026-09-03","note":"450 million parameters"},{"field":"paper_url,release_date","url":"https://arxiv.org/abs/2506.01844","source_type":"literature","fetched":"2026-09-03","note":"2025-06-02 提交"},{"field":"release_date,weights_url","url":"https://huggingface.co/blog/smolvla","source_type":"hub","fetched":"2026-09-03","note":"June 3, 2025; lerobot/smolvla_base"},{"field":"code_url","url":"https://github.com/huggingface/lerobot","source_type":"official","fetched":"2026-09-03","note":"LeRobot 代码 Apache 2.0"},{"field":"params_b,release_date,weights_url,license","url":"https://huggingface.co/api/models/lerobot/smolvla_base","source_type":"hub","fetched":"2026-09-04","note":"safetensors 450,046,176；createdAt 2025-06-01；模型卡无 license 字段"}]},{"id":"openvla","name":"OpenVLA","org":"Stanford / UC Berkeley / TRI 等（OpenVLA 团队）","release_date":"2024-06","params_b":7.54,"arch_type":"monolithic_autoregressive","arch_notes":"Prismatic VLM：DINOv2 + SigLIP 双视觉编码器 + Llama-2 7B，动作离散为 token 自回归输出","license":"MIT（代码与权重仓库标注）；权重衍生自 Llama-2，受 Llama 2 Community License 约束","commercial_ok":null,"weights_url":"https://huggingface.co/openvla/openvla-7b","code_url":"https://github.com/openvla/openvla","paper_url":"https://arxiv.org/abs/2406.09246","modalities_in":["rgb","language"],"action_space":"7-DoF 末端位姿增量 + 夹爪（离散 token）","cross_embodiment":"Open X-Embodiment 约 97 万条轨迹预训练","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"license,release_date,code_url","url":"https://github.com/openvla/openvla","source_type":"official","fetched":"2026-09-04","note":"README：All code is made available under the MIT License；Models derive from Llama-2, subject to the Llama Community License；首发 2024-06-13"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/openvla/openvla-7b","source_type":"hub","fetched":"2026-09-04","note":"safetensors 7,541,237,184；模型卡 license: mit；createdAt 2024-06-10"},{"field":"paper_url,cross_embodiment","url":"https://arxiv.org/abs/2406.09246","source_type":"literature","fetched":"2026-09-04","note":"OpenVLA: An Open-Source Vision-Language-Action Model；970K trajectories from Open X-Embodiment"}]},{"id":"x-vla","name":"X-VLA","org":"清华大学 AIR（2toINF）","release_date":"2025-10","params_b":0.88,"arch_type":"soft_prompted_transformer","arch_notes":"Soft-Prompted Transformer，Florence-2-large 视觉；0.9B 规模；ICLR 2026","license":"Apache-2.0（代码与模型卡）","commercial_ok":true,"weights_url":"https://huggingface.co/2toINF/X-VLA-Pt","code_url":"https://github.com/2toinf/X-VLA","paper_url":"https://arxiv.org/abs/2510.10274","modalities_in":["rgb","language","state"],"action_space":"跨本体软提示，单臂/双臂","cross_embodiment":"CALVIN / LIBERO / VLABench / RoboTwin2 / BridgeDataV2；Franka / AgiBot G1 / Google Robot / WidowX / AgileX","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm"],"evidence":[{"field":"license,release_date,cross_embodiment,code_url","url":"https://github.com/2toinf/X-VLA","source_type":"official","fetched":"2026-09-04","note":"README：Apache License 2.0；arXiv 2510.10274（2025-10）；ICLR 2026；Franka / Agibot-G1 / Google Robot / WidowX / Agilex"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/2toINF/X-VLA-Pt","source_type":"hub","fetched":"2026-09-04","note":"safetensors 879,738,545；license: apache-2.0；base microsoft/Florence-2-large"}]},{"id":"octo","name":"Octo","org":"UC Berkeley RAIL 等","release_date":"2024-05","params_b":0.093,"arch_type":"transformer_policy","arch_notes":"Transformer 主干 + 扩散动作头；octo-base 93M、octo-small 27M","license":"MIT（代码与权重仓库标注）","commercial_ok":true,"weights_url":"https://huggingface.co/rail-berkeley/octo-base-1.5","code_url":"https://github.com/octo-models/octo","paper_url":"https://arxiv.org/abs/2405.12213","modalities_in":["rgb","language","goal_image"],"action_space":"末端位姿增量 + 夹爪（扩散）","cross_embodiment":"Open X-Embodiment 约 80 万条轨迹预训练","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"params_b,license,code_url,cross_embodiment","url":"https://github.com/octo-models/octo","source_type":"official","fetched":"2026-09-04","note":"README：Octo-Base 93M Params / Octo-Small 27M；License: MIT；800k robot trajectories from Open X-Embodiment"},{"field":"weights_url,license,release_date","url":"https://huggingface.co/api/models/rail-berkeley/octo-base-1.5","source_type":"hub","fetched":"2026-09-04","note":"模型卡 license: mit；createdAt 2024-05-21"}]},{"id":"rdt-1b","name":"RDT-1B","org":"清华大学 TSAIL（thu-ml）","release_date":"2024-10","params_b":1,"arch_type":"transformer_policy","arch_notes":"扩散 Transformer；输入语言 + 最多三路 RGB，一次预测 64 步动作；另有 RDT-170M 小版本","license":"MIT（代码、权重、数据均为 MIT，README 原文）","commercial_ok":true,"weights_url":"https://huggingface.co/robotics-diffusion-transformer/rdt-1b","code_url":"https://github.com/thu-ml/RoboticsDiffusionTransformer","paper_url":"https://arxiv.org/abs/2410.07864","modalities_in":["rgb","language","state"],"action_space":"统一动作空间，覆盖单臂/双臂、关节/末端、位置/速度","cross_embodiment":"100 万+ 多机器人 episodes 预训练，ALOHA 双臂 6K+ episodes 微调","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm","mobile_manipulator"],"evidence":[{"field":"license,params_b,cross_embodiment,code_url","url":"https://github.com/thu-ml/RoboticsDiffusionTransformer","source_type":"official","fetched":"2026-09-04","note":"README：All the code, model weights, and data are licensed under MIT license；1B-parameter；1M+ multi-robot episodes"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/robotics-diffusion-transformer/rdt-1b","source_type":"hub","fetched":"2026-09-04","note":"模型卡 license: mit；createdAt 2024-08-27"},{"field":"paper_url,release_date","url":"https://arxiv.org/abs/2410.07864","source_type":"literature","fetched":"2026-09-04","note":"RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation，2024-10"}]},{"id":"cogact","name":"CogACT","org":"Microsoft Research","release_date":"2024-12","params_b":null,"arch_type":"vlm_plus_action_expert","arch_notes":"VLM 主干（prism-dinosiglip-224px）+ DiT 扩散动作模块（DiT-S/B/L 三档）；一次输出 16 步 7-DoF 动作","license":"MIT（代码、权重、数据均为 MIT，README 原文）","commercial_ok":true,"weights_url":"https://huggingface.co/CogACT/CogACT-Base","code_url":"https://github.com/microsoft/CogACT","paper_url":"https://arxiv.org/abs/2411.19650","modalities_in":["rgb","language"],"action_space":"7-DoF 末端位姿增量 + 夹爪（扩散，16 步动作块）","cross_embodiment":"Open X-Embodiment 预训练","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"license,release_date,arch_notes,code_url","url":"https://github.com/microsoft/CogACT","source_type":"official","fetched":"2026-09-04","note":"README：released 2024-12-01；All the code, model weights, and data are licensed under MIT license；CogACT-Small/Base/Large，DiT-S/B/L"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/CogACT/CogACT-Base","source_type":"hub","fetched":"2026-09-04","note":"模型卡 license: mit；createdAt 2024-11-29"}]},{"id":"spatialvla","name":"SpatialVLA","org":"上海人工智能实验室 等","release_date":"2025-01","params_b":4.03,"arch_type":"monolithic_autoregressive","arch_notes":"PaliGemma2-3B 主干 + 空间位置编码与自适应动作网格；1.1M 真机 episodes 预训练","license":"MIT（代码 README 与模型卡均标 MIT）","commercial_ok":true,"weights_url":"https://huggingface.co/IPEC-COMMUNITY/spatialvla-4b-224-pt","code_url":"https://github.com/SpatialVLA/SpatialVLA","paper_url":"https://arxiv.org/abs/2501.15830","modalities_in":["rgb","language"],"action_space":"末端位姿增量 + 夹爪（自适应离散网格）","cross_embodiment":"Open X-Embodiment + RH20T 共 1.1M episodes","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,cross_embodiment,code_url","url":"https://github.com/SpatialVLA/SpatialVLA","source_type":"official","fetched":"2026-09-04","note":"README：2025-01-29 发布；MIT license；1.1 Million real robot episodes from Open X-Embodiment and RH20T；Google Robot / WidowX / Franka"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/IPEC-COMMUNITY/spatialvla-4b-224-pt","source_type":"hub","fetched":"2026-09-04","note":"safetensors 4,027,854,731；license: MIT；base_model google/paligemma2-3b-pt-224"}]},{"id":"univla","name":"UniVLA","org":"OpenDriveLab（香港大学 等）","release_date":"2025-05","params_b":7.54,"arch_type":"vlm_plus_action_expert","arch_notes":"Prismatic 7B 主干 + 任务中心潜在动作（VQ-VAE 两阶段）+ 约 12M 参数动作解码器","license":"Apache-2.0（代码仓库与模型卡均标注）","commercial_ok":true,"weights_url":"https://huggingface.co/qwbu/univla-7b","code_url":"https://github.com/OpenDriveLab/UniVLA","paper_url":"https://arxiv.org/abs/2505.06111","modalities_in":["rgb","language"],"action_space":"潜在动作 → 各本体解码","cross_embodiment":"Open X-Embodiment + Ego4D 人类视频子集；LIBERO / CALVIN / SimplerEnv / 导航 / AgileX 真机","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm","wheeled"],"evidence":[{"field":"license,release_date,arch_notes,cross_embodiment,code_url","url":"https://github.com/OpenDriveLab/UniVLA","source_type":"official","fetched":"2026-09-04","note":"README：Apache-2.0；2025-05 代码发布；7B backbone，action decoder around 12M；OXE + Ego4D；AgiLex 真机"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/qwbu/univla-7b","source_type":"hub","fetched":"2026-09-04","note":"safetensors 7,541,237,184；license: apache-2.0；createdAt 2025-05-06"},{"field":"paper_url","url":"https://arxiv.org/abs/2505.06111","source_type":"literature","fetched":"2026-09-04","note":"UniVLA: Learning to Act Anywhere with Task-centric Latent Actions"}]},{"id":"molmoact","name":"MolmoAct","org":"Allen Institute for AI（Ai2）","release_date":"2025-08","params_b":8.12,"arch_type":"monolithic_autoregressive","arch_notes":"Molmo 系：SigLIP2 视觉 + Qwen2.5-7B 语言 + VQVAE 深度编码，先推理空间轨迹再出动作","license":"Apache-2.0（代码与权重）","commercial_ok":true,"weights_url":"https://huggingface.co/allenai/MolmoAct-7B-D-0812","code_url":"https://github.com/allenai/MolmoAct","paper_url":"https://arxiv.org/abs/2508.07917","modalities_in":["rgb","depth","language"],"action_space":"末端位姿增量 + 夹爪","cross_embodiment":"MolmoAct 自采 1 万 episodes（LeRobot 格式）+ 预训练混合","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,cross_embodiment,code_url","url":"https://github.com/allenai/MolmoAct","source_type":"official","fetched":"2026-09-04","note":"README：2025-08-12 模型发布、2025-10-24 代码发布；Apache-2.0；10k robot episodes 自采；Google Robot / Franka"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/allenai/MolmoAct-7B-D-0812","source_type":"hub","fetched":"2026-09-04","note":"safetensors 8,119,417,808（F32）；license: apache-2.0；base Qwen/Qwen2.5-7B + siglip2-so400m"}]},{"id":"rynnvla-001","name":"RynnVLA-001","org":"阿里巴巴达摩院","release_date":"2025-08","params_b":7.01,"arch_type":"villa","arch_notes":"以第一人称视频生成模型为主干（Chameleon 系），先做视频生成预训练再接轨迹建模与动作","license":"Apache-2.0（代码与模型卡）；README 同时注明 research preview intended for non-commercial use ONLY","commercial_ok":null,"weights_url":"https://huggingface.co/Alibaba-DAMO-Academy/RynnVLA-001-7B-Base","code_url":"https://github.com/alibaba-damo-academy/RynnVLA-001","paper_url":"https://arxiv.org/abs/2509.15212","modalities_in":["rgb","language"],"action_space":"机械臂关节/末端（LeRobot 格式数据）","cross_embodiment":"第一人称人类视频预训练迁移到机械臂","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,arch_notes,code_url","url":"https://github.com/alibaba-damo-academy/RynnVLA-001","source_type":"official","fetched":"2026-09-04","note":"README：2025-08-08 发布；Apache 2.0；research preview intended for non-commercial use ONLY"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/Alibaba-DAMO-Academy/RynnVLA-001-7B-Base","source_type":"hub","fetched":"2026-09-04","note":"safetensors 7,013,158,912；license: apache-2.0；base chameleon"}]},{"id":"wall-oss","name":"WALL-OSS","org":"自变量机器人（X Square Robot）","release_date":"2025-09","params_b":4.22,"arch_type":"vlm_plus_action_expert","arch_notes":"Qwen2.5-VL 主干 + MoE 动作专家（Qwen2_5_VLMoEForAction）；FLOW / FAST 两种动作头；2026-05 发布 0.5 版与世界模型 WALL-WM","license":"Apache-2.0（代码仓库）；权重模型卡未标注许可证","commercial_ok":null,"weights_url":"https://huggingface.co/x-square-robot/wall-oss-flow","code_url":"https://github.com/X-Square-Robot/wall-x","paper_url":"https://arxiv.org/abs/2509.11766","modalities_in":["rgb","language","state"],"action_space":"双臂关节（flow matching / FAST token）","cross_embodiment":"大规模多模态预训练 + 真机操作（README 未列本体清单）","pretrain_hours":null,"target_embodiments":["dual_arm"],"evidence":[{"field":"release_date,license,arch_notes,code_url","url":"https://github.com/X-Square-Robot/wall-x","source_type":"official","fetched":"2026-09-04","note":"README：2025-09 WALL-OSS；2026-05 Wall-OSS-0.5 与 WALL-WM；Apache-2.0 license；Qwen2_5_VLMoEForAction"},{"field":"params_b,weights_url","url":"https://huggingface.co/api/models/x-square-robot/wall-oss-flow","source_type":"hub","fetched":"2026-09-04","note":"safetensors 4,224,041,072；模型卡无 license 字段；base qwen2_5_vl"}]},{"id":"internvla-m1","name":"InternVLA-M1","org":"上海人工智能实验室 InternRobotics","release_date":"2025-09","params_b":3.75,"arch_type":"vlm_plus_action_expert","arch_notes":"Qwen2.5-VL-3B 主干，空间引导 + 多模态多任务联合训练","license":"代码 MIT；权重模型卡标 CC BY-NC-SA 4.0","commercial_ok":false,"weights_url":"https://huggingface.co/InternRobotics/InternVLA-M1","code_url":"https://github.com/InternRobotics/InternVLA-M1","paper_url":"https://arxiv.org/abs/2510.13778","modalities_in":["rgb","language"],"action_space":"末端位姿增量 + 夹爪","cross_embodiment":"RT-1 / Bridge 预训练，LIBERO 与 SimplerEnv 微调","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"license,arch_notes,code_url","url":"https://github.com/InternRobotics/InternVLA-M1","source_type":"official","fetched":"2026-09-04","note":"README：MIT license（代码）；Qwen2.5-VL backbone；0918 Release model weights"},{"field":"params_b,weights_url,license,release_date","url":"https://huggingface.co/api/models/InternRobotics/InternVLA-M1","source_type":"hub","fetched":"2026-09-04","note":"safetensors 3,754,622,976；license: cc-by-nc-sa-4.0；createdAt 2025-09-16；base Qwen2.5-VL-3B-Instruct"},{"field":"paper_url","url":"https://arxiv.org/abs/2510.13778","source_type":"literature","fetched":"2026-09-04","note":"InternVLA-M1 技术报告"}]},{"id":"graspvla","name":"GraspVLA","org":"北京大学 EPIC 实验室 / 银河通用","release_date":"2025-07","params_b":null,"arch_type":"vlm_plus_action_expert","arch_notes":"InternLM2-1.8B 语言 + DINOv2/SigLIP 视觉；十亿帧合成抓取数据 SynGrasp-1B 预训练","license":"CC BY-NC 4.0（代码徽章与模型卡一致）","commercial_ok":false,"weights_url":"https://huggingface.co/vegebirrd/GraspVLA","code_url":"https://github.com/PKU-EPIC/GraspVLA","paper_url":"https://arxiv.org/abs/2505.03233","modalities_in":["rgb","language"],"action_space":"抓取位姿 / 末端轨迹","cross_embodiment":"合成数据 SynGrasp-1B（240 类、1 万+ 物体）","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,arch_notes,cross_embodiment,code_url","url":"https://github.com/PKU-EPIC/GraspVLA","source_type":"official","fetched":"2026-09-04","note":"README：2025-07-25 模型发布；CC BY-NC 4.0；internlm2-1_8b + DINOv2/SigLIP；SynGrasp-1B"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/vegebirrd/GraspVLA","source_type":"hub","fetched":"2026-09-04","note":"模型卡 license: cc-by-nc-4.0"}]},{"id":"being-h0","name":"Being-H0","org":"BeingBeyond（北京）","release_date":"2025-08","params_b":8.11,"arch_type":"monolithic_autoregressive","arch_notes":"InternVL 系 VLA，显式手部运动建模 + GRVQ 运动分词器；1B / 8B / 14B 三档","license":"MIT（代码仓库与模型卡）","commercial_ok":true,"weights_url":"https://huggingface.co/BeingBeyond/Being-H0-8B-2508","code_url":"https://github.com/BeingBeyond/Being-H0","paper_url":"https://arxiv.org/abs/2507.15597","modalities_in":["rgb","language"],"action_space":"手部关节运动（灵巧手）","cross_embodiment":"大规模人类手部视频预训练，后训练对齐机器人","pretrain_hours":null,"target_embodiments":["dexterous_hand"],"evidence":[{"field":"release_date,license,arch_notes,code_url","url":"https://github.com/BeingBeyond/Being-H0","source_type":"official","fetched":"2026-09-04","note":"README：2025-08-02 代码与模型发布；MIT license；1B/8B/14B；dexterous hands"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/BeingBeyond/Being-H0-8B-2508","source_type":"hub","fetched":"2026-09-04","note":"safetensors 8,114,397,181；license: MIT"}]},{"id":"eo-1","name":"EO-1","org":"IPEC-COMMUNITY（上海人工智能实验室 系）","release_date":"2025-08","params_b":3.77,"arch_type":"monolithic_autoregressive","arch_notes":"单一解码器 Transformer，离散自回归 + 连续 flow matching 统一；Qwen2.5-VL-3B 主干；EO-Data1.5M 交错数据","license":"MIT（模型卡）","commercial_ok":true,"weights_url":"https://huggingface.co/IPEC-COMMUNITY/EO-1-3B","code_url":"https://github.com/IPEC-PUBLIC/EO-1","paper_url":"https://arxiv.org/abs/2508.21112","modalities_in":["rgb","language","state"],"action_space":"关节/末端（flow matching）","cross_embodiment":"AgiBotWorld + Open X-Embodiment + RoboMIND + SO100 社区数据；Franka Panda / WidowX 250S / AgiBot G-1 / SO100","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm"],"evidence":[{"field":"arch_notes,cross_embodiment,code_url","url":"https://github.com/IPEC-PUBLIC/EO-1","source_type":"official","fetched":"2026-09-04","note":"README：Qwen2.5-VL-3B-Instruct；EO-Data1.5M；Franka Panda, WidowX 250 S, AgiBot G-1, LeRobot SO100"},{"field":"params_b,weights_url,license,release_date","url":"https://huggingface.co/api/models/IPEC-COMMUNITY/EO-1-3B","source_type":"hub","fetched":"2026-09-04","note":"safetensors 3,771,607,072；license: MIT；createdAt 2025-08-28"},{"field":"paper_url","url":"https://arxiv.org/abs/2508.21112","source_type":"literature","fetched":"2026-09-04","note":"EO-1: An Open Unified Embodied Foundation Model for General Robot Control"}]},{"id":"lingbot-vla-4b","name":"LingBot-VLA 4B","org":"蚂蚁灵波（Robbyant）","release_date":"2026-01","params_b":4.2,"arch_type":"vlm_plus_action_expert","arch_notes":"Qwen2.5-VL-3B-Instruct 主干 + MoGe-2 深度组件；有 depth-free 与 depth-distilled 两版；2 万小时真机双臂数据","license":"Apache-2.0（代码仓库）；权重模型卡未标注许可证","commercial_ok":null,"weights_url":"https://huggingface.co/robbyant/lingbot-vla-4b","code_url":"https://github.com/robbyant/lingbot-vla","paper_url":"https://arxiv.org/abs/2601.18692","modalities_in":["rgb","language","state"],"action_space":"双臂关节","cross_embodiment":"9 种双臂构型 2 万小时真机数据；AgiBot G1 / AgileX / Galaxea R1Pro 实测","pretrain_hours":20000,"target_embodiments":["dual_arm","dual_arm_wheeled"],"evidence":[{"field":"release_date,license,arch_notes,cross_embodiment,pretrain_hours,code_url","url":"https://github.com/robbyant/lingbot-vla","source_type":"official","fetched":"2026-09-04","note":"README：2026-01-27 技术报告与权重代码发布；Apache-2.0 License；Qwen2.5-VL-3B-Instruct；20,000 hours of real-world data from 9 dual-arm configurations；Agibot G1 / AgileX / Galaxea R1Pro"},{"field":"params_b,weights_url","url":"https://huggingface.co/api/models/robbyant/lingbot-vla-4b","source_type":"hub","fetched":"2026-09-04","note":"safetensors 4,197,425,739；模型卡无 license 字段；createdAt 2026-01-26"}]},{"id":"nora","name":"NORA","org":"Declare Lab（新加坡科技设计大学）","release_date":"2025-04","params_b":3.76,"arch_type":"monolithic_autoregressive","arch_notes":"Qwen2.5-VL-3B 主干，动作 token 自回归；有 NORA-1.5 与 NORA-LONG 变体","license":"仓库与模型卡均未标注许可证","commercial_ok":null,"weights_url":"https://huggingface.co/declare-lab/nora","code_url":"https://github.com/declare-lab/nora","paper_url":"https://arxiv.org/abs/2504.19854","modalities_in":["rgb","language"],"action_space":"7-DoF（6 + 夹爪）","cross_embodiment":"Open X-Embodiment 混合预训练；WidowX / BridgeData V2 评测","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"arch_notes,cross_embodiment,code_url","url":"https://github.com/declare-lab/nora","source_type":"official","fetched":"2026-09-04","note":"README：built on Qwen2.5 VL；OXE mixture；WidowX / BridgeData V2；7-DoF"},{"field":"params_b,weights_url,release_date","url":"https://huggingface.co/api/models/declare-lab/nora","source_type":"hub","fetched":"2026-09-04","note":"safetensors 3,758,262,272；模型卡无 license 字段；createdAt 2025-04-28"}]},{"id":"vla-adapter","name":"VLA-Adapter","org":"OpenHelix 团队","release_date":"2025-09","params_b":0.5,"arch_type":"vlm_plus_action_expert","arch_notes":"Prismatic 架构 + Qwen2.5-0.5B 语言主干，轻量策略头（Pro 版策略约 207MB）","license":"MIT（代码仓库）；权重模型卡未标注","commercial_ok":null,"weights_url":"https://huggingface.co/VLA-Adapter/LIBERO-Spatial","code_url":"https://github.com/OpenHelix-Team/VLA-Adapter","paper_url":"https://arxiv.org/abs/2509.09372","modalities_in":["rgb","language"],"action_space":"末端位姿增量 + 夹爪","cross_embodiment":"LIBERO / CALVIN 微调；ALOHA（Cobot Magic）真机部署","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm"],"evidence":[{"field":"release_date,license,params_b,arch_notes,cross_embodiment,code_url","url":"https://github.com/OpenHelix-Team/VLA-Adapter","source_type":"official","fetched":"2026-09-04","note":"README：论文 2025-09-11、代码 2025-09-22；MIT license；0.5B；Qwen2.5-0.5B；LIBERO / CALVIN；ALOHA / Cobot Magic"}]},{"id":"hpt","name":"HPT（Heterogeneous Pre-trained Transformers）","org":"MIT（Lirui Wang, Kaiming He 等）","release_date":"2024-09","params_b":0.227,"arch_type":"transformer_policy","arch_notes":"把不同本体对齐到共享潜空间的 Transformer 主干；XLarge 226.8M / Large 50.5M / Base 12.6M / Small 3.1M；NeurIPS 2024 Spotlight","license":"MIT（代码仓库）；模型卡未标注许可证","commercial_ok":null,"weights_url":"https://huggingface.co/liruiw/hpt-xlarge","code_url":"https://github.com/liruiw/HPT","paper_url":"https://arxiv.org/abs/2409.20537","modalities_in":["rgb","state","language"],"action_space":"各本体自有动作头","cross_embodiment":"多本体数据混合预训练（README 未列具体清单）","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm"],"evidence":[{"field":"params_b,license,arch_notes,code_url","url":"https://github.com/liruiw/HPT","source_type":"official","fetched":"2026-09-05","note":"README：HPT-XLarge 226.8M / Large 50.5M / Base 12.6M / Small 3.1M；MIT license；NeurIPS 2024 Spotlight"},{"field":"weights_url","url":"https://huggingface.co/api/models/liruiw/hpt-xlarge","source_type":"hub","fetched":"2026-09-05","note":"模型卡无 license 字段；createdAt 2024-05-15"}]},{"id":"lapa","name":"LAPA","org":"KAIST / 华盛顿大学 等（Latent Action Pretraining）","release_date":"2024-11","params_b":7,"arch_type":"monolithic_autoregressive","arch_notes":"从无动作标注视频学潜在动作再预训练 VLA；基于 LWM-Chat-1M（Large World Model）；ICLR 2025","license":"MIT（代码与模型卡均标注）","commercial_ok":true,"weights_url":"https://huggingface.co/latent-action-pretraining/LAPA-7B-openx","code_url":"https://github.com/LatentActionPretraining/LAPA","paper_url":"https://arxiv.org/abs/2410.11758","modalities_in":["rgb","language"],"action_space":"潜在动作 → 微调后映射真实动作","cross_embodiment":"Open X-Embodiment 预训练；SIMPLER 评测","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,cross_embodiment,code_url","url":"https://github.com/LatentActionPretraining/LAPA","source_type":"official","fetched":"2026-09-05","note":"README：weights released 2024-11-22；MIT License；Open-X Embodiment；ICLR 2025"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/latent-action-pretraining/LAPA-7B-openx","source_type":"hub","fetched":"2026-09-05","note":"模型卡 license: mit；base_model LargeWorldModel/LWM-Chat-1M-Jax"}]},{"id":"llava-vla","name":"LLaVA-VLA","org":"OpenHelix 团队","release_date":"2025-06","params_b":7,"arch_type":"monolithic_autoregressive","arch_notes":"直接基于 LLaVA-v1.5-7B / LLaVA-OneVision 0.5B 的 VLA；ICRA 2026","license":"MIT（代码与模型卡）","commercial_ok":true,"weights_url":"https://huggingface.co/chenpyyy/LLaVA-VLA","code_url":"https://github.com/OpenHelix-Team/LLaVA-VLA","paper_url":"https://arxiv.org/abs/2602.22663","modalities_in":["rgb","language"],"action_space":"末端位姿增量 + 夹爪","cross_embodiment":"CALVIN 微调；RoboTwin 评测","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,arch_notes,code_url","url":"https://github.com/OpenHelix-Team/LLaVA-VLA","source_type":"official","fetched":"2026-09-05","note":"README：2025-06-17 发布代码与权重、2025-07-05 发 0.5B；MIT license；基于 LLaVA"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/chenpyyy/LLaVA-VLA","source_type":"hub","fetched":"2026-09-05","note":"模型卡 license: MIT；createdAt 2025-06-16"}]},{"id":"robovlms","name":"RoboVLMs（KosMos P.H.）","org":"清华大学 / 字节跳动 Research / 中科院自动化所 等","release_date":"2024-12","params_b":null,"arch_type":"vlm_plus_action_expert","arch_notes":"把任意 VLM 变成 VLA 的框架；表现最好的是 KosMos 主干 + 策略头；Nature Machine Intelligence 2026","license":"Apache-2.0（代码与模型卡）","commercial_ok":true,"weights_url":"https://huggingface.co/robovlms/RoboVLMs","code_url":"https://github.com/Robot-VLAs/RoboVLMs","paper_url":"https://arxiv.org/abs/2412.14058","modalities_in":["rgb","language"],"action_space":"末端位姿增量 + 夹爪","cross_embodiment":"Open X-Embodiment 预训练；CALVIN / SimplerEnv / 真机","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,arch_notes,code_url","url":"https://github.com/Robot-VLAs/RoboVLMs","source_type":"official","fetched":"2026-09-05","note":"README：initial release 12/11/24；Apache-2.0 license；KosMos VLM backbone；Nature Machine Intelligence"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/robovlms/RoboVLMs","source_type":"hub","fetched":"2026-09-05","note":"模型卡 license: apache-2.0；checkpoints kosmos_ph_oxe-pretrain.pt 等"}]},{"id":"magma","name":"Magma-8B","org":"Microsoft Research","release_date":"2025-02","params_b":8.9,"arch_type":"monolithic_autoregressive","arch_notes":"Llama-3-8B-Instruct + ConvNeXt 视觉；Set-of-Mark / Trace-of-Mark 预训练；同时做 UI 导航与机器人操作","license":"MIT（代码与模型卡）","commercial_ok":true,"weights_url":"https://huggingface.co/microsoft/Magma-8B","code_url":"https://github.com/microsoft/Magma","paper_url":"https://arxiv.org/abs/2502.13130","modalities_in":["rgb","language"],"action_space":"7-DoF 末端位姿增量 + 夹爪","cross_embodiment":"Open X-Embodiment + Ego4D 等；WidowX / Bridge、LIBERO 评测","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,arch_notes,cross_embodiment,code_url","url":"https://github.com/microsoft/Magma","source_type":"official","fetched":"2026-09-05","note":"README：2025-02-25 发布；MIT；LLama-3-8B-Instruct + ConvNeXt；7 DoF；WidowX/Bridge、LIBERO"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/microsoft/Magma-8B","source_type":"hub","fetched":"2026-09-05","note":"safetensors 8,903,066,496；license: MIT"}]},{"id":"vjepa2-ac","name":"V-JEPA 2-AC","org":"Meta FAIR","release_date":"2025-06","params_b":1,"arch_type":"villa","arch_notes":"自监督视频世界模型 V-JEPA 2（ViT-g 1B）之上加动作条件预测器，通过规划实现零样本抓取/放置；权重在 HF 上为受限（gated）访问","license":"MIT 为主，部分文件 Apache-2.0（README 原文）；HF 权重需申请访问","commercial_ok":null,"weights_url":"https://huggingface.co/collections/facebook/v-jepa-2-6841bad8413014e185b497a6","code_url":"https://github.com/facebookresearch/vjepa2","paper_url":"https://arxiv.org/abs/2506.09985","modalities_in":["rgb","state"],"action_space":"末端位姿（基于图像目标的规划）","cross_embodiment":"在 Franka 上做 reaching / grasping / pick-and-place 规划","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,arch_notes,cross_embodiment,code_url","url":"https://github.com/facebookresearch/vjepa2","source_type":"official","fetched":"2026-09-05","note":"README：V-JEPA 2 2025-06-25；majority licensed under MIT, portions under separate terms；ViT-g/16 1B；Franka arm planning"},{"field":"weights_url","url":"https://huggingface.co/api/models/facebook/vjepa2-ac-vitg-384","source_type":"hub","fetched":"2026-09-05","note":"模型卡 API 返回 401（受限访问）"}]},{"id":"dexvla","name":"DexVLA","org":"美的 AI 研究院 等（DexVLA 团队）","release_date":"2025-02","params_b":3,"arch_type":"vlm_plus_action_expert","arch_notes":"Qwen2-VL-2B 主干 + 插件式 ScaleDP 扩散专家（1B 或 410M）","license":"代码 MIT；ScaleDP 预训练权重模型卡标 CC BY-NC 4.0","commercial_ok":false,"weights_url":"https://huggingface.co/lesjie/scale_dp_h","code_url":"https://github.com/juruobenruo/DexVLA","paper_url":"https://arxiv.org/abs/2502.05855","modalities_in":["rgb","language","state"],"action_space":"双臂 / 灵巧手关节（扩散）","cross_embodiment":"ALOHA 脚本、AgileX 评测；README 未列训练小时数","pretrain_hours":null,"target_embodiments":["dual_arm","dexterous_hand"],"evidence":[{"field":"release_date,arch_notes,license,code_url","url":"https://github.com/juruobenruo/DexVLA","source_type":"official","fetched":"2026-09-05","note":"README：2025-02-17 发布；Qwen2-VL-2B + 1B ScaleDP（另有 410M）；MIT LICENSE 文件"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/lesjie/scale_dp_h","source_type":"hub","fetched":"2026-09-05","note":"模型卡 license: cc-by-nc-4.0；createdAt 2025-02-24"}]},{"id":"gr-1","name":"GR-1","org":"字节跳动 Research","release_date":"2023-12","params_b":null,"arch_type":"villa","arch_notes":"GPT 式模型：输入语言 + 观测序列 + 状态，同时预测动作与未来帧；大规模视频生成式预训练；ICLR 2024","license":"Apache-2.0（代码仓库）","commercial_ok":null,"weights_url":"https://lf-robot-opensource.bytetos.com/obj/lab-robot-public/gr1_code_release/","code_url":"https://github.com/bytedance/GR-1","paper_url":"https://arxiv.org/abs/2312.13139","modalities_in":["rgb","language","state"],"action_space":"末端位姿增量 + 夹爪","cross_embodiment":"CALVIN 微调（ABCD / ABC 两组权重）","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,license,arch_notes,weights_url,code_url","url":"https://github.com/bytedance/GR-1","source_type":"official","fetched":"2026-09-05","note":"README：ICLR 2024；Apache-2.0 License；权重 snapshot_ABCD.pt / snapshot_ABC.pt 在字节对象存储；GPT-style video generative pre-training"}]},{"id":"spirit-v1.5","name":"Spirit v1.5","org":"千寻智能（Spirit AI）","release_date":"2026-01","params_b":5.4,"arch_type":"vlm_plus_action_expert","arch_notes":"Qwen3-VL 主干 + DiT 动作头 + policy API；2026-01-11 起在 RoboChallenge Table30 榜单排名第一（站方自述）","license":"代码 MIT（README）；模型卡标 apache-2.0","commercial_ok":true,"weights_url":"https://huggingface.co/Spirit-AI-robotics/Spirit-v1.5","code_url":"https://github.com/Spirit-AI-Team/spirit-v1.5","paper_url":"https://www.spirit-ai.com/en/blog/spirit-v1-5","modalities_in":["rgb","language","state"],"action_space":"动作块（chunk）输出","cross_embodiment":"Table30 基准任务；README 未列本体清单与训练小时","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm"],"evidence":[{"field":"release_date,license,arch_notes,code_url","url":"https://github.com/Spirit-AI-Team/spirit-v1.5","source_type":"official","fetched":"2026-09-05","note":"README：2026-01 发布；MIT；Qwen3-VL backbone + DiT head；#1 on RoboChallenge Table30 as of Jan 11, 2026"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/Spirit-AI-robotics/Spirit-v1.5","source_type":"hub","fetched":"2026-09-05","note":"safetensors 5,400,901,236（F32）；license: apache-2.0；createdAt 2026-01-11"}]},{"id":"galaxea-g0.5","name":"Galaxea G0.5","org":"星海图（Galaxea）","release_date":"2026-06","params_b":2,"arch_type":"monolithic_autoregressive","arch_notes":"Qwen3.5 2B VLM，单一自回归流同时输出推理与动作；前代 G0Plus 3B / G0Tiny 250M（2026-01）","license":"G0.5 Community License（非商用 + 有限专利许可；模型卡 license: other）；2026-01-04 前版本为 Apache-2.0","commercial_ok":false,"weights_url":"https://huggingface.co/OpenGalaxea/G05","code_url":"https://github.com/OpenGalaxea/GalaxeaVLA","paper_url":"https://arxiv.org/abs/2608.11739","modalities_in":["rgb","language","state"],"action_space":"移动操作（底盘 + 双臂）","cross_embodiment":"14 种本体；R1 Lite / R1 Pro / SO-100/101 / DROID-Franka；Galaxea Open-World Dataset 500+ 小时","pretrain_hours":500,"target_embodiments":["dual_arm_wheeled","single_arm"],"evidence":[{"field":"release_date,license,arch_notes,cross_embodiment,pretrain_hours,code_url","url":"https://github.com/OpenGalaxea/GalaxeaVLA","source_type":"official","fetched":"2026-09-05","note":"README：G0.5 checkpoint 2026-06-16、论文 2026-08-12；G0.5 Community License (Non-Commercial + Limited Patent License)；Qwen3.5 2B；14 embodiments；500+ hours"},{"field":"weights_url,license","url":"https://huggingface.co/api/models/OpenGalaxea/G05","source_type":"hub","fetched":"2026-09-05","note":"模型卡 license: other (G05-Community-License)；createdAt 2026-06-16"}]},{"id":"rdt2-vq","name":"RDT2-VQ","org":"清华大学 TSAIL（thu-ml）","release_date":"2025-09","params_b":8.29,"arch_type":"monolithic_autoregressive","arch_notes":"Qwen2.5-VL-7B-Instruct 改造的自回归 VLA，Residual VQ 动作分词器；另有 flow-matching 版 RDT2-FM；以 UMI 手持夹爪采的人类操作数据预训练，零样本跨本体","license":"Apache-2.0（代码与模型卡）","commercial_ok":true,"weights_url":"https://huggingface.co/robotics-diffusion-transformer/RDT2-VQ","code_url":"https://github.com/thu-ml/RDT2","paper_url":"https://arxiv.org/abs/2602.03310","modalities_in":["rgb","language"],"action_space":"24 步 × 20 维相对动作块（双臂末端）","cross_embodiment":"10,000+ 小时、100+ 室内场景的 UMI 人类操作数据；双臂 UR5e / 双臂 Franka FR3 零样本","pretrain_hours":10000,"target_embodiments":["dual_arm"],"evidence":[{"field":"release_date,license,arch_notes,cross_embodiment,pretrain_hours,code_url","url":"https://github.com/thu-ml/RDT2","source_type":"official","fetched":"2026-09-05","note":"README：2025-09 模型、2026-02 论文；Apache-2.0；Qwen2.5-VL-7B-Instruct；Residual VQ；10,000+ hours；Bimanual UR5e / FR3 / UMI gripper"},{"field":"params_b,weights_url,license","url":"https://huggingface.co/api/models/robotics-diffusion-transformer/RDT2-VQ","source_type":"hub","fetched":"2026-09-05","note":"safetensors 8,292,166,656；license: apache-2.0；createdAt 2025-09-22"}]},{"id":"internvla-a1.5","name":"InternVLA-A1.5","org":"上海人工智能实验室 InternRobotics","release_date":"2026-07","params_b":2.69,"arch_type":"vlm_plus_action_expert","arch_notes":"Qwen3.5-2B 原生 VLM + 轻量统一动作专家；base / RoboTwin / LIBERO / DOMINO 四个权重","license":"CC BY-NC-SA 4.0（代码仓库与模型卡一致）","commercial_ok":false,"weights_url":"https://huggingface.co/InternRobotics/InternVLA-A1.5-base","code_url":"https://github.com/InternRobotics/InternVLA-A1","paper_url":"https://arxiv.org/abs/2607.04988","modalities_in":["rgb","language","state"],"action_space":"关节 / 末端（动作专家）","cross_embodiment":"InternData-A1 + LeRobot v2.1 数据；RoboTwin 2.0 / LIBERO / SimplerEnv / DOMINO","pretrain_hours":null,"target_embodiments":["single_arm","dual_arm"],"evidence":[{"field":"license,arch_notes,cross_embodiment,code_url","url":"https://github.com/InternRobotics/InternVLA-A1","source_type":"official","fetched":"2026-09-05","note":"README：CC BY-NC-SA 4.0；Qwen3.5-2B backbone + lightweight unified action expert；InternData-A1；RoboTwin 2.0 / LIBERO / SimplerEnv / DOMINO"},{"field":"params_b,weights_url,license,release_date","url":"https://huggingface.co/api/models/InternRobotics/InternVLA-A1.5-base","source_type":"hub","fetched":"2026-09-05","note":"safetensors 2,685,264,547；license: cc-by-nc-sa-4.0；createdAt 2026-07-04；base Qwen/Qwen3.5-2B"}]},{"id":"vla-0","name":"VLA-0","org":"NVIDIA Research","release_date":"2025-10","params_b":3,"arch_type":"monolithic_autoregressive","arch_notes":"不改架构、把动作直接当文本让 Qwen2.5-VL-3B-Instruct 输出；无大规模机器人预训练","license":"CC BY-NC 4.0（代码与权重）；底座受 Qwen Research License 约束","commercial_ok":false,"weights_url":"https://huggingface.co/ankgoyal/vla0-libero","code_url":"https://github.com/NVlabs/vla0","paper_url":"https://arxiv.org/abs/2510.13054","modalities_in":["rgb","language"],"action_space":"末端位姿增量 + 夹爪（文本形式输出）","cross_embodiment":"LIBERO 仿真；SO-100 真机","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"license,arch_notes,cross_embodiment,code_url","url":"https://github.com/NVlabs/vla0","source_type":"official","fetched":"2026-09-05","note":"README：CC BY-NC 4.0 license；base model subject to Qwen Research License；Qwen2.5-VL-3B-Instruct；LIBERO；SO-100 robot"},{"field":"weights_url","url":"https://huggingface.co/api/models/ankgoyal/vla0-libero","source_type":"hub","fetched":"2026-09-05","note":"模型卡 lastModified 2025-11-18；无 license 字段与参数计数"}]},{"id":"minivla","name":"MiniVLA","org":"Stanford ILIAD","release_date":"2024-12","params_b":0.5,"arch_type":"monolithic_autoregressive","arch_notes":"OpenVLA 缩小版：Prismatic 架构 + Qwen2.5-0.5B 语言主干 + DINOSigLIP 视觉，256 个额外动作 token","license":"代码 MIT；README 注明预训练权重可能继承底座（Llama Community License）限制","commercial_ok":null,"weights_url":"https://huggingface.co/collections/Stanford-ILIAD/minivla-675a2a9aca369ff3a6c04e33","code_url":"https://github.com/Stanford-ILIAD/openvla-mini","paper_url":"https://arxiv.org/abs/2406.09246","modalities_in":["rgb","language"],"action_space":"7-DoF 末端位姿增量 + 夹爪（离散 token）","cross_embodiment":"Open X-Embodiment 混合；LIBERO 四套任务；WidowX / BridgeData V2","pretrain_hours":null,"target_embodiments":["single_arm"],"evidence":[{"field":"release_date,params_b,license,arch_notes,cross_embodiment,code_url,weights_url","url":"https://github.com/Stanford-ILIAD/openvla-mini","source_type":"official","fetched":"2026-09-05","note":"README：2024-12-09 Added MiniVLA configs with Qwen2.5 0.5B backbone；All code is made available under the MIT License；pretrained models may inherit restrictions from the underlying base models（Llama Community License）；LIBERO / WidowX"}]},{"id":"genie-envisioner","name":"Genie Envisioner（GE-Act）","org":"智元机器人（AgiBot）","release_date":"2025-08","params_b":null,"arch_type":"villa","arch_notes":"视频世界模型 GE-Base（LTX-Video 系）+ 动作模型 GE-Act + 仿真器 GE-Sim（Cosmos2）；以 AgiBotWorld 数据训练","license":"代码分两块：LTX/Cosmos/pipeline/openpi_client 部分 Apache-2.0，其余代码与数据 CC BY-NC-SA 4.0；权重模型卡未标注","commercial_ok":null,"weights_url":"https://huggingface.co/agibot-world/Genie-Envisioner","code_url":"https://github.com/AgibotTech/Genie-Envisioner","paper_url":"https://arxiv.org/abs/2508.05635","modalities_in":["rgb","language"],"action_space":"关节（GE-Act，基于视频潜空间）","cross_embodiment":"AgiBotWorld 数据集（README 未量化小时）","pretrain_hours":null,"target_embodiments":["dual_arm_wheeled"],"evidence":[{"field":"release_date,license,arch_notes,cross_embodiment,code_url","url":"https://github.com/AgibotTech/Genie-Envisioner","source_type":"official","fetched":"2026-09-05","note":"README：论文 2025-08-08、代码 08-13、权重 08-14；Apache-2.0（部分目录）+ CC BY-NC-SA 4.0（其余代码与数据）；LTX-Video / Cosmos2；AgiBotWorld"},{"field":"weights_url","url":"https://huggingface.co/api/models/agibot-world/Genie-Envisioner","source_type":"hub","fetched":"2026-09-05","note":"模型卡 Genie-Envisioner-v1.0：GE_base_fast_v0.1 / ge_base_slow_v0.1 safetensors；无 license 字段；createdAt 2025-08-13"}]},{"id":"gr00t-n1.5-3b","name":"GR00T N1.5 3B","org":"NVIDIA","release_date":"2025-05","params_b":2.72,"arch_type":"dual_system","arch_notes":"SigLIP2 视觉 + T5 语言编码 + flow matching 动作 Transformer（adaptive layernorm）；按本体 ID 索引可变维本体感知；已被 N1.6 / N1.7 接替","license":"模型卡标 Nvidia License（页面注明仅限非商用）","commercial_ok":false,"weights_url":"https://huggingface.co/nvidia/GR00T-N1.5-3B","code_url":"https://github.com/NVIDIA/Isaac-GR00T","paper_url":"https://arxiv.org/abs/2503.14734","modalities_in":["rgb","language","state"],"action_space":"关节（flow matching 动作块）","cross_embodiment":"跨本体模型，人形 / 双臂 / 单臂（模型卡未列具体机型）","pretrain_hours":null,"target_embodiments":["humanoid","dual_arm","single_arm"],"evidence":[{"field":"params_b,release_date,weights_url","url":"https://huggingface.co/api/models/nvidia/GR00T-N1.5-3B","source_type":"hub","fetched":"2026-09-05","note":"safetensors 2,724,163,520；createdAt 2025-05-28"},{"field":"license,arch_notes","url":"https://huggingface.co/nvidia/GR00T-N1.5-3B","source_type":"hub","fetched":"2026-09-05","note":"模型卡：Nvidia License（Non-commercial use only）；SigLip2 + T5 + flow matching transformer；cross-embodiment"},{"field":"code_url","url":"https://github.com/NVIDIA/Isaac-GR00T","source_type":"official","fetched":"2026-09-05","note":"Isaac-GR00T 代码仓库（Apache-2.0）"}]}]}