Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics
这是一份面向零基础读者的结构化研究笔记。本文只记录公开论文文本、arXiv 元数据和作者公开链接能支持的结论;本站没有复现医疗机器人训练、仿真或真实手术实验,因此不会把论文报告的成功率写成本站 E4 结果。
一句话讲什么(TL;DR)
Open-H-Embodiment 是一个面向 medical robotics 的大规模开放数据集项目。论文报告它包含 119 个 datasets、780 小时 paired video and kinematic data,来自 50+ institutions,覆盖 20 个 healthcare robot platforms、33 task families 和 5 environment types。它的目标是解决医疗机器人 foundation model 缺数据、数据不开放、单本体、单机构的问题。
论文还展示了这个数据集能支撑两个 downstream foundation models:GR00T-H,一个医疗机器人 VLA;Cosmos-H-Surgical-Simulator,一个 action-conditioned surgical world model。GR00T-H 在 SutureBot end-to-end suturing benchmark 中完成 5/20 trials,也就是 25%,而对照模型为 0%;在 29-step ex vivo suturing sequence 中达到约 64% average success。
如果只记一个直觉:Open-H 不是一篇“又一个模型”的论文,而是在给医疗机器人补“公共训练食材”。没有足够开放、多机构、多机器人、多任务的数据,再强模型也很难学会手术场景。
所以这一节是想说:Open-H 的核心贡献是医疗机器人开放数据基础设施。
这是个什么场景
医疗机器人,尤其手术机器人,有很强的现实价值:减少医生负担、提高精度、扩大医疗服务覆盖。但它也是最难做自动化的机器人场景之一。场景里有软组织、出血、器械遮挡、细小动作、严格安全边界,还涉及隐私和伦理。
通用机器人领域已经有 Open X-Embodiment、DROID、AgiBot 等大规模数据集,推动了 VLA 和 generalist robot policy。但 surgical robotics 长期缺少类似数据基础。很多数据集很小,只来自单个医院、单个机器人、单个任务,很难训练跨平台 foundation model。
Open-H 的场景就是补这个缺口。它想把全球多个机构、多种医疗机器人、多种手术和医疗操作数据组织成统一格式,让研究者可以训练和评估医疗机器人 foundation models。
医疗机器人数据困境
小数据集 单机构 单机器人 不开放
│ │ │ │
└──────────┴──────────┴──────────┘
▼
foundation model 难以泛化
▼
Open-H: 多机构 + 多本体 + 多任务 + 开放格式
医疗机器人和普通桌面机械臂不同。普通 pick-and-place 任务失败可能只是没拿到物体;手术场景失败可能伤害组织、延长手术时间或影响患者安全。因此,数据集不仅要大,还要有清楚边界和可追溯来源。
所以这一节是想说:医疗机器人最缺的是可共享、可训练、可比较的大规模数据。

之前的人怎么做的,为什么不够好
过去的 surgical robotics 数据集很多是小而专的。论文提到 JIGSAWS 长期是重要 benchmark,但只有约 3 小时 da Vinci demonstrations。SutureBot 提供约 6 小时 end-to-end suturing data,ImitateCholec 有约 20 小时 segmented cholecystectomy demonstrations。这些数据都很有价值,但规模和多样性不足。
另一个问题是 single-embodiment。一个数据集只覆盖 da Vinci 或 dVRK,不代表模型能迁移到 Versius、MIRA、BiTrack、Maestro 或其他平台。医疗机器人硬件差异很大:相机、器械、自由度、控制接口、任务流程都不同。
第三个问题是开放性。医疗数据涉及隐私、合规、机构限制,很多数据很难公开共享。没有开放数据,社区就无法像通用机器人那样共同训练、共同比较、共同复现。
论文也指出,通用 VLA 在 surgical task 上并不天然强。SutureBot benchmark 曾评估 OpenVLA、GR00T-N1、π0 等 general-purpose VLA,结果它们被从 scratch 训练的 multitask ACT policy 明显超过。这说明“通用模型”不自动拥有手术技能,医疗领域需要自己的数据和后训练。
所以这一节是想说:医疗机器人不是缺模型名字,而是缺足够开放、足够多样的训练数据。
这篇论文的新想法
第一,新想法是 Open-H-Embodiment 数据集本身。论文报告数据集由 119 datasets 构成,总计 780 小时 paired video and kinematic data,来自 50+ institutions,覆盖 20 distinct robot platforms、33 task families、5 environment types。
第二,新想法是跨本体医疗机器人数据。它不只收 da Vinci,也包含 CMR Versius、dVRK、Rob Surgical BiTrack、Virtual Incision MIRA、Moon Surgical Maestro、UR5e、Franka Panda、Kuka Med 14、custom systems 和 simulation 等。
第三,新想法是用数据集训练和验证 foundation models。论文展示 GR00T-H 和 Cosmos-H-Surgical-Simulator,分别对应 policy / VLA 方向和 world model / simulation 方向。
第四,新想法是把数据贡献流程和模型材料一起公开。论文提到 GitHub、Hugging Face dataset、GR00T-H weights、Cosmos-H-Surgical-Simulator weights 等材料入口。对社区来说,数据能不能被使用和持续贡献,和模型结果同样重要。
Open-H 论文结构
Open-H Dataset
├─ 119 datasets / 780 hours
├─ 50+ institutions
├─ 20 healthcare robot platforms
└─ 33 task families / 5 environment types
Downstream models
├─ GR00T-H: surgical VLA / policy
└─ Cosmos-H-Surgical-Simulator: action-conditioned world model
所以这一节是想说:Open-H 把数据、模型和社区贡献流程绑在一起。
它分几步做的(方法)
第 1 步:汇聚多机构医疗机器人数据
输入是来自多个机构、多个机器人平台、多个医疗任务的数据。每份数据可能有不同相机、不同机器人状态、不同任务标签、不同采样率和不同格式。
处理过程是统一整理成 paired video and kinematic data。paired 的意思是视频和机器人运动状态有对应关系,不只是单独的视频。kinematic data 可以理解为机器人关节、器械位置、动作轨迹等运动信息。
输出是一个可用于训练 foundation model 的数据 corpus。它不是只给人看视频,而是给模型学习“看到什么动作”和“机器人实际怎么动”的对应关系。
第 2 步:覆盖不同机器人本体和任务族
Open-H 把 robot platforms 分成多类。论文报告 20 个 healthcare robot platforms,覆盖 surgical robotic systems、industrial arms modified for healthcare、flexible endoscope robots、simulated robots、manual instrumentation 等。
任务也不是单一 suturing。它覆盖 prostatectomy、cholecystectomy、hysterectomy、hernia repair 等完整 procedure,也覆盖 suturing、knot tying、tissue manipulation、needle handover、needle pickup 等 subtask。
输入是多种任务粒度;处理是把 procedure-level 和 subtask-level 数据都纳入;输出是一个能支持从短技能到长流程学习的数据集。
第 3 步:训练 GR00T-H
GR00T-H 是 healthcare-focused foundational VLA。论文说明它基于 NVIDIA GR00T-N1.6,并在 Open-H dataset 上 post-training。
输入是医疗机器人数据中的视频、状态和任务提示。处理是后训练,让通用 foundation policy 适配 surgical domain。输出是能在 suturing、handover、knot tying 等任务中执行或预测动作的 policy。
在 SutureBot end-to-end suturing benchmark 上,GR00T-H 是唯一完成 full end-to-end task completion 的模型,完成 5/20 trials,也就是 25%。论文还报告在 ex vivo suturing sequence 中达到 64% average success。
第 4 步:评估数据效率和跨本体泛化
论文不只看最终成功率,还看 fine-tuning data efficiency。它比较 33% data 和 100% data 下 GR00T-H、ACT、GR00T-N1.6 的表现。论文报告 33% data 时 GR00T-H 已经和 ACT 在 3-task average 上接近,full data 时更明显领先。
跨本体方面,论文在 CMR Versius、Virtual Incision MIRA、dVRK-Si 等平台上验证。输入是不同 robot embodiment;处理是用 Open-H post-training 作为更强初始化;输出是跨平台任务表现提升。
第 5 步:训练 Cosmos-H-Surgical-Simulator
Cosmos-H-Surgical-Simulator 是 action-conditioned world model。它基于 Cosmos-Predict 2.5,在 Open-H surgical data mixture 上 fine-tune。
输入是视频帧和 kinematic action trajectory;处理是预测未来视频或生成 surgical simulation rollout;输出是一个能跨九个 robotic platforms 的 surgical video world model。
这一步的意义是:医疗机器人不只需要 policy,还需要能模拟动作后果的世界模型。世界模型可以支持 in silico policy evaluation、synthetic data generation 和 training screening。
第 6 步:公开材料和贡献流程
论文强调数据集、模型权重、训练和推理代码等材料入口。对数据集论文来说,这一步不是附属内容,而是核心方法的一部分。只有材料可获取、格式可复用、贡献流程可执行,Open-H 才能成为基础设施。
不过,公开不等于无风险。医疗机器人数据仍涉及隐私、许可证、机构治理、数据偏差和临床安全边界。使用者必须遵守数据集和模型各自的 license 与合规要求。
所以这一节是想说:Open-H 的方法是“数据汇聚 -> 格式统一 -> 模型后训练 -> 跨本体评估 -> 世界模型 -> 开放生态”。

关键数字
| 数字 | 原文语境 | 这说明什么 |
|---|---|---|
| 119 | datasets | 数据不是单一来源,而是多集合汇聚 |
| 780 hours | paired video and kinematic data | 规模远大于传统小型 surgical robotics 数据集 |
| 50+ | participating institutions | 多机构降低单一医院偏差 |
| 20 | healthcare robot platforms | 支撑 cross-embodiment 研究 |
| 33 | task families | 覆盖多种医疗机器人任务 |
| 5 | environment types | 从 simulation 到 live clinical procedures |
| 25% | GR00T-H SutureBot end-to-end completion | 5/20 trials,其他对照为 0% |
| 64% | 29-step ex vivo suturing average success | 论文报告的 clinically proximate evaluation |
| 9 | Cosmos-H 跨 robotic platforms | world model 从单一平台扩展到多本体 |
| 65,000 | GR00T-H training steps | 论文补充材料里的训练设置 |
这些数字全部是论文报告,不是本站复现实验。尤其 25% 和 64% 只能写作 paper-reported performance,不能写成本站验证。
所以这一节是想说:Open-H 的证据重点是数据规模、跨本体覆盖和两个 downstream model 的可行性。
实验结果说明了什么
实验首先说明大规模医疗机器人数据确实能改善 foundation policy。GR00T-H 在 SutureBot 中完成 end-to-end task,而对照模型没有完成;在 out-of-distribution wound configuration 和 lighting change 下也有更好的平均表现。这支持一个判断:通用模型需要医疗领域 post-training 才能处理 surgical domain 的长程任务。
其次,实验说明 Open-H 可以提升 data efficiency。使用更少 fine-tuning data 时,GR00T-H 已经接近或超过一些 baseline;full data 时进一步提升。这对医疗机器人很重要,因为真实标注数据昂贵,不能指望每个医院都收大量本地数据。
第三,Cosmos-H-Surgical-Simulator 说明 Open-H 不只服务 policy,也服务 world model。一个 action-conditioned surgical world model 可以把动作和未来视觉变化连接起来,这对仿真评估、合成数据和训练筛选都有价值。
但实验也暴露边界。25% end-to-end suturing completion 对研究很重要,但离临床可靠性还很远。64% ex vivo subtask average success 说明模型在接近真实组织的环境里有信号,但仍不能用于临床安全声明。
所以这一节是想说:实验支持 Open-H 作为研究基础设施,但远未证明医疗机器人自动化已经可临床部署。
术语表
- Medical robotics:医疗场景中的机器人,包括手术、超声、内窥镜和辅助操作。
- Paired video and kinematic data:视频和机器人运动状态成对记录的数据。
- Cross-embodiment:跨不同机器人平台或身体结构。
- VLA:Vision-Language-Action model,把视觉、语言和动作连接起来。
- GR00T-H:基于 GR00T-N1.6 后训练的医疗机器人 foundation VLA。
- SutureBot:自主缝合 benchmark。
- Ex vivo:离体组织实验,比如猪皮组织,不是人体临床部署。
- World model:预测动作对未来观察影响的模型。
- Action-conditioned:以动作轨迹作为条件生成或预测未来状态。
所以这一节是想说:Open-H 的关键词围绕医疗数据、跨本体和手术世界模型。
局限和边界
第一,医疗机器人数据有天然合规边界。即使论文公开数据和模型入口,也不意味着所有数据都能任意商业或临床使用。
第二,数据规模大不等于任务均衡。不同平台、手术类型、机构和任务时长可能分布不均,模型可能偏向高频平台或高频 procedure。
第三,benchmark 成功不等于临床安全。25% end-to-end completion 和 64% ex vivo average success 是研究进展,但不满足临床可靠性要求。
第四,world model 生成视频不等于真实物理完全正确。动作条件视频预测可能视觉合理,但力、接触、组织变形和安全边界仍要单独验证。
第五,跨本体泛化仍受硬件接口影响。不同机器人自由度、控制延迟、器械结构和视觉系统差异很大,不是一个统一 dataset 就能完全消除。
所以这一节是想说:Open-H 是基础设施突破,但医疗机器人落地仍有强合规和安全门槛。
和其他论文的关系
和 mimo-embodied 相比,Open-H 更像数据底座。MiMo 证明跨具身模型可以连接驾驶和机器人,Open-H 证明医疗机器人需要自己的跨本体开放数据。
和 alanavlm 相比,Open-H 关注动作和 kinematics,AlanaVLM 关注 egocentric video understanding。一个更接近 policy training,另一个更接近 embodied perception。
和 embodied-3d-generation-survey 相比,Open-H 提供真实医疗机器人数据,3D generation survey 讨论如何生成 simulation-ready assets 和 environments。两者都服务模型训练,只是一个来自真实数据,一个来自生成和仿真。
和 open-x-embodiment 相比,Open-H 可以看作医疗机器人领域的专门化版本,但医疗数据更敏感、任务更精细、合规约束更强。
所以这一节是想说:Open-H 是 Batch 6 里最典型的数据基础设施论文。
和本导读的关系
本站前面已经有 Open X-Embodiment、DROID、LeRobot、AgiBot 等数据和工具链笔记。Open-H 把这条线推进到医疗机器人,说明“foundation model 需要开放数据”这个规律在高风险领域同样成立,但实现难度更高。
它适合放在 dataset-eval 主题里,也适合和 VLA、world-model、sim 三类笔记交叉阅读。读者应该重点理解:医疗机器人模型的瓶颈不只是模型架构,而是数据、格式、许可证、机构协作和安全评估。
所以这一节是想说:Open-H 补齐本站医疗机器人 foundation data 的关键一块。
思考题
- 为什么医疗机器人比普通桌面机械臂更需要多机构数据?
- Paired video and kinematic data 为什么比单纯手术视频更有训练价值?
- GR00T-H 的 25% end-to-end completion 应该如何解读,为什么不能夸大?
- Cosmos-H-Surgical-Simulator 和 GR00T-H 分别解决什么问题?
- 如果一个医院想用 Open-H fine-tune 自己的机器人,需要额外验证哪些风险?
FAQ
Q:Open-H 是不是一个模型? A:核心是数据集和开放生态。论文也展示 GR00T-H 和 Cosmos-H-Surgical-Simulator,但它的基础贡献是数据。
Q:780 小时是不是全部真实手术视频? A:不是。论文说覆盖 5 environment types,从 digital simulation 到 live clinical procedures。具体组成要看原文和数据说明。
Q:GR00T-H 25% 成功率是不是已经能做手术? A:不是。25% 是 SutureBot benchmark 中的研究结果,临床部署需要远高于此的可靠性、安全验证和监管审批。
Q:本站有没有下载或训练 Open-H? A:没有。这里只记录论文报告和公开来源,不做本地训练或医疗数据处理。
进一步读什么
- Open X-Embodiment:理解通用机器人开放数据如何推动 VLA。
- LeRobot:理解机器人数据格式和训练工具链。
- SutureBot:理解自主缝合 benchmark。
- Cosmos / world model 笔记:理解 action-conditioned simulator 的意义。
后续核验清单
如果之后要把本文从 UNVERIFIED 提升到人工核验状态,应逐项核对:119 datasets、780 hours、50+ institutions、20 platforms、33 task families、5 environment types;GR00T-H 是否基于 GR00T-N1.6;SutureBot 5/20 trials = 25%;ex vivo 29-step sequence 64%;Cosmos-H-Surgical-Simulator 是否跨 9 platforms;公开材料链接和 license 是否按原文表述。
原文信息
- arXiv: 2604.21017
- PDF: https://arxiv.org/pdf/2604.21017
- Project: https://open-h.github.io/open-h-embodiment/
- Repository: https://github.com/open-h/open-h-embodiment
@article{nelson2026openh,
title = {Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics},
author = {{Open-H-Embodiment Consortium} and Nelson, Nigel and Chen, Juo-Tung and Haworth, Jesse and Chen, Xinhao and others},
journal = {arXiv preprint arXiv:2604.21017},
year = {2026}
}
◼
引用本笔记 / Cite this note
@online{eai_open_h_embodiment_2026,
title = {(readable note) Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics},
author = {Xun, Jason},
year = {2026},
note = {Note on a 2026 paper},
howpublished = {\url{https://estelledc.github.io/embodied-ai-reading-station/papers/open-h-embodiment/}},
organization = {Embodied AI: Zero to One}
}
All 202 papers (full index)
- 1. LLaVA: Visual Instruction Tuning
- 2. 3DShape2VecSet: 3D Shape Representation for Diffusion Models
- 3. SayCan: Do As I Can, Not As I Say
- 4. OpenVLA: An Open-Source Vision-Language-Action Model
- 5. VLAS: VLA Model With Speech Instructions
- 6. MLA: Multisensory Language-Action Model
- 7. Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control
- 8. CartoRadar: RF-Based 3D SLAM Rivaling Vision Approaches
- 9. mmCLIP: Boosting mmWave-based Zero-shot HAR via Signal-Text Alignment
- 10. mmNorm: Non-Line-of-Sight 3D Object Reconstruction via mmWave Surface Normal Estimation
- 11. Proactive Hearing Assistants that Isolate Egocentric Conversations
- 12. NeuralAids: Wireless Hearables With Programmable Speech AI Accelerators
- 13. Creating speech zones with self-distributing acoustic swarms
- 14. Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation
- 15. SoundStream: An End-to-End Neural Audio Codec
- 16. AudioLM
- 17. Conformer
- 18. Dual-path RNN
- 19. EnCodec
- 20. Meta-StyleSpeech
- 21. MusicLM
- 22. Robust Speech Recognition via Large-Scale Weak Supervision
- 23. SeamlessM4T
- 24. Stable Audio
- 25. Universal Source Separation with Weakly Labelled Data
- 26. Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning
- 27. RLBench: The Robot Learning Benchmark & Learning Environment
- 28. robosuite: A Modular Simulation Framework and Benchmark for Robot Learning
- 30. CALVIN
- 31. LIBERO
- 32. RH20T
- 33. What Matters in Learning from Offline Human Demonstrations for Robot Manipulation
- 34. DROID
- 35. Open X-Embodiment
- 36. RoboCasa
- 37. SimplerEnv
- 38. Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
- 39. 3D Diffusion Policy: Generalizable Visuomotor Policy Learning via Simple 3D Representations
- 40. Consistency Policy: Accelerated Visuomotor Policies via Consistency Distillation
- 41. EquiBot: SIM(3)-Equivariant Diffusion Policy
- 42. DiT-Policy
- 43. Diffusion Policy Policy Optimization (DPPO)
- 44. Affordance-based Robot Manipulation with Flow Matching
- 45. FlowPolicy: 3D Flow-based Policy via Consistency Flow Matching
- 46. FAST: Efficient Action Tokenization for VLA
- 47. π₀: A Vision-Language-Action Flow Model for General Robot Control
- 48. pi_0.5: VLA with Open-World Generalization
- 49. A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning
- 50. Generative Adversarial Imitation Learning
- 51. Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (ACT/ALOHA)
- 52. AnyTeleop
- 53. Behavior Transformers: Cloning k Modes with One Stone
- 54. Implicit Behavioral Cloning
- 55. RoboCat
- 56. ALOHA 2
- 58. HumanPlus
- 59. Generalizable Humanoid Manipulation with 3D Diffusion Policies (iDP3)
- 60. Mobile ALOHA
- 61. SmolVLA
- 62. Universal Manipulation Interface
- 63. Behavior Generation with Latent Actions (VQ-BeT)
- 64. ImageBind: One Embedding Space To Bind Them All
- 65. Connecting Touch and Vision via Cross-Modal Prediction
- 66. AnyMAL: An Efficient and Scalable Any-Modality Augmented Language Model
- 67. AudioPaLM
- 68. FROMAGe: Grounding LLMs to Images
- 69. OneLLM
- 70. X-VLM: Multi-Grained Vision Language Pre-Training
- 71. Tactile Beyond Pixels (Sparsh-X)
- 72. Sparsh: Self-supervised Touch Representations
- 73. Tactile-VLA
- 74. TLA: Tactile-Language-Action
- 75. Code as Policies: Language Model Programs for Embodied Control
- 76. Inner Monologue: Embodied Reasoning through Planning with Language Models
- 77. LLM+P: Empowering LLMs with Optimal Planning
- 78. PaLM-E: An Embodied Multimodal Language Model
- 79. ProgPrompt
- 80. ChatGPT for Robotics
- 81. GenSim
- 82. RoboFlamingo
- 83. Tree-Planner
- 84. VoxPoser
- 85. See Through Smoke: Robust Indoor Mapping with Low-cost mmWave Radar
- 86. Can WiFi Estimate Person Pose?
- 87. 3DRIMR: 3D Reconstruction and Imaging via mmWave Radar based on Deep Learning
- 88. milliEgo: Single-chip mmWave Radar Aided Egomotion Estimation via Deep Sensor Fusion
- 89. High Resolution Point Clouds from mmWave Radar
- 90. RadarSLAM: Radar based Large-Scale SLAM in All Weathers
- 91. Through-Wall Pose Imaging in Real-Time with a Many-to-Many Encoder/Decoder Paradigm
- 92. RFMask: A Simple Baseline for Human Silhouette Segmentation with Radio Signals
- 93. RFPose-OT: RF-Based 3D Human Pose Estimation via Optimal Transport Theory
- 94. Argus: Multi-View Egocentric Human Mesh Reconstruction Based on Stripped-Down Wearable mmWave Add-on
- 95. Diffusion Model is a Good Pose Estimator from 3D RF-Vision
- 96. Enabling Visual Recognition at Radio Frequency (PanoRadar)
- 97. Wave-Former: Through-Occlusion 3D Reconstruction via Wireless Shape Completion
- 98. Habitat: A Platform for Embodied AI Research
- 99. Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning
- 101. Habitat 2.0
- 102. ManiSkill
- 103. ProcTHOR
- 104. SAPIEN: A SimulAted Part-based Interactive ENvironment
- 105. BEHAVIOR-1K
- 106. BridgeData V2
- 106. Habitat 3.0
- 107. Isaac Lab
- 108. DexMV
- 108. MuJoCo Playground
- 109. DexCap
- 109. RT-1: Robotics Transformer for Real-World Control at Scale
- 110. 3D Diffusion Policy (DP3)
- 111. Octo: An Open-Source Generalist Robot Policy
- 112. RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control
- 113. RT-Trajectory: Robotic Task Generalization via Hindsight Trajectory Sketches
- 114. 3D-VLA
- 116. GR-2: Generative Video-Language-Action Model
- 117. DexVLA
- 117. OpenHelix
- 118. Cosmos World Foundation Model
- 118. OpenVLA-OFT
- 119. RDT-1B: Diffusion Foundation Model for Bimanual Manipulation
- 120. RoboMamba
- 121. SpatialVLA
- 122. TinyVLA
- 123. TraceVLA: Visual Trace Prompting
- 124. Learning Transferable Visual Models From Natural Language Supervision
- 125. Flamingo: a Visual Language Model for Few-Shot Learning
- 126. BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
- 127. BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation
- 128. DeepSeek-VL: Towards Real-World Vision-Language Understanding
- 129. EVA-CLIP: Improved Training Techniques for CLIP at Scale
- 130. FILIP: Fine-grained Interactive Language-Image Pre-Training
- 131. Florence-2: Advancing a Unified Representation for a Variety of Vision Tasks
- 132. InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks
- 133. Improved Baselines with Visual Instruction Tuning
- 134. OBELICS
- 135. Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond
- 136. Sigmoid Loss for Language Image Pre-Training
- 137. What matters when building vision-language models?
- 138. Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
- 139. The Llama 3 Herd of Models
- 140. LLaVA-NeXT-Interleave
- 141. LLaVA-OneVision: Easy Visual Task Transfer
- 142. Long-CLIP: Unlocking the Long-Text Capability of CLIP
- 143. Pixtral 12B
- 144. Dream to Control: Learning Behaviors by Latent Imagination
- 145. World Models
- 146. DayDreamer
- 147. Mastering Atari with Discrete World Models
- 148. Dreamer V3: Mastering Diverse Domains through World Models
- 149. Transformers are Sample-Efficient World Models
- 150. TWM: Transformer-based World Models
- 151. 1X World Model Challenge
- 153. GAIA-1
- 154. Genie: Generative Interactive Environments
- 155. Navigation World Models
- 156. UniSim
- 157. LeRobot: An Open-Source Library for End-to-End Robot Learning
- 158. CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulation
- 159. Universal Actions for Enhanced Embodied Foundation Models
- 160. LoHoVLA: A Unified Vision-Language-Action Model for Long-Horizon Embodied Tasks
- 161. AutoRT: Embodied Foundation Models for Large Scale Orchestration of Robotic Agents
- 162. EO-1: Interleaved Vision-Text-Action Pretraining for General Robot Control
- 163. Qwen-VLA: Unifying Vision-Language-Action Modeling across Tasks, Environments, and Robot Embodiments
- 164. RealMirror: A Comprehensive, Open-Source Vision-Language-Action Platform for Embodied AI
- 165. LLaDA-VLA: Vision Language Diffusion Action Models
- 166. Discrete Diffusion VLA: Bringing Discrete Diffusion to Action Decoding in Vision-Language-Action Policies
- 167. Vlaser: Vision-Language-Action Model with Synergistic Embodied Reasoning
- 168. X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model
- 169. Embodiment Transfer Learning for Vision-Language-Action Models
- 170. HiMoE-VLA: Hierarchical Mixture-of-Experts for Generalist Vision-Language-Action Policies
- 171. Green-VLA: Staged Vision-Language-Action Model for Generalist Robots
- 172. AC^2-VLA: Action-Context-Aware Adaptive Computation in Vision-Language-Action Models for Efficient Robotic Manipulation
- 173. MoS-VLA: A Vision-Language-Action Model with One-Shot Skill Adaptation
- 174. Asynchronous Fast-Slow Vision-Language-Action Policies for Whole-Body Robotic Manipulation
- 175. VLA-Forget: Vision-Language-Action Unlearning for Embodied Foundation Models
- 176. Membership Inference Attacks on Vision-Language-Action Models
- 177. A Survey on Efficient Vision-Language-Action Models
- 178. Survey of Vision-Language-Action Models for Embodied Manipulation
- 179. Embodied AI with Foundation Models for Mobile Service Robots: A Systematic Review
- 180. Toward Embodied AGI: A Review of Embodied AI and the Road Ahead
- 181. RoboNeuron: A Middle-Layer Infrastructure for Agent-Driven Orchestration in Embodied AI
- 182. Embodied Navigation Foundation Model
- 183. MiMo-Embodied: X-Embodied Foundation Model Technical Report
- 184. Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics
- 185. AlanaVLM: A Multimodal Embodied AI Foundation Model for Egocentric Video Understanding
- 186. 3D Generation for Embodied AI and Robotic Simulation: A Survey
- 187. DISCO: Language-Guided Manipulation with Diffusion Policies and Constrained Inpainting
- 188. Time-Unified Diffusion Policy with Action Discrimination for Robotic Manipulation
- 189. Learning Diffusion Policy from Primitive Skills for Robot Manipulation
- 190. Trace-Focused Diffusion Policy for Multi-Modal Action Disambiguation in Long-Horizon Robotic Manipulation
- 191. Gaze2Act: Gaze-Conditioned Vision-Language-Action Policies for Interactive Robot Manipulation
- 192. LACY: A Vision-Language Model-based Language-Action Cycle for Self-Improving Robotic Manipulation
- 193. villa-X: Enhancing Latent Action Modeling in Vision-Language-Action Models
- 194. InstructVLA: Vision-Language-Action Instruction Tuning from Understanding to Manipulation
- 195. Discrete Policy: Learning Disentangled Action Space for Multi-Task Robotic Manipulation
- 196. Towards Generalizable Vision-Language Robotic Manipulation: A Benchmark and LLM-guided 3D Policy
- 197. A Survey of Language-Conditioned Robot Manipulation
- 198. SafeEmbodAI: a Safety Framework for Mobile Robots in Embodied AI Systems
- 199. The Essential Role of Causality in Foundation World Models for Embodied AI
- 200. A call for embodied AI
- 201. Toward General-Purpose Robots via Foundation Models: A Survey and Meta-Analysis
- 202. What Breaks Embodied AI Security: LLM Vulnerabilities, CPS Flaws, or Something Else?