3D Generation for Embodied AI and Robotic Simulation: A Survey
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 元数据和项目页能支持的结论;本站没有复现任何 3D 生成、仿真或 sim-to-real 实验,因此不会把 survey 中的方向判断写成本站实验结论。
一句话讲什么(TL;DR)
这篇 survey 讨论 3D generation 如何服务 embodied AI and robotic simulation。它不是普通“3D 生成技术综述”,而是从机器人需要什么出发,把 3D generation 分成三个角色:Data Generator、Simulation Environments、Sim2Real Bridge。
论文的核心判断是:embodied AI 需要的 3D 生成不只是“看起来真实”。生成物体要有 articulation、mass、friction、material、kinematics,生成场景要支持 interaction and task execution,生成世界还要能帮助 robot learning 从 simulation 转到 real world。字段里出现 URDF、MJCF、USD、GLB、physics parameters、digital twin 等,都是因为机器人要在仿真器里真正动起来。
如果只记一个直觉:普通 3D 生成像“画漂亮家具”,embodied 3D generation 要生成“门能打开、杯子能抓、布会变形、机器人能在里面训练”的世界。
所以这一节是想说:本文把 3D 生成从视觉逼真推进到交互和仿真可用。
这是个什么场景
机器人学习很依赖仿真。真实世界训练成本高、慢、危险;仿真可以并行生成大量任务、失败不会损坏设备,也方便控制变量。但仿真世界要有足够多样、真实、可交互的 3D assets 和 environments。
传统 3D generation 关注外观:形状像不像、纹理好不好、渲染是否真实。但机器人不只看外观。机器人要推门,需要门有铰链;要拖布,需要布能变形;要抓杯子,需要杯子有可接触几何;要移动椅子,需要质量、摩擦和碰撞体。
这篇 survey 的问题是:3D generation 怎样从“视觉内容生成”变成“具身训练基础设施”?它把整个领域整理成 object-level asset creation、scene-level environment synthesis、simulation-to-reality transfer 三条线。
普通 3D 生成 vs 具身 3D 生成
普通目标:外观逼真
├─ mesh / texture / image quality
└─ 人看起来像真的
具身目标:交互可用
├─ articulation / kinematics
├─ mass / friction / material
├─ URDF / MJCF / simulator compatibility
└─ robot can perceive, plan, act, and learn
这个场景和本站的 simulation、world model、dataset-eval 都有关。3D generation 是给机器人造训练世界,simulation 是让世界运行起来,world model 是学习世界变化,policy learning 是在世界里学动作。
所以这一节是想说:embodied AI 里的 3D 生成必须服务动作和物理,而不只是服务渲染。

之前的人怎么做的,为什么不够好
已有 3D generation surveys 很多,但多按模型技术组织,比如 NeRF、3DGS、diffusion、VAE、GAN、text-to-3D、image-to-3D。这样的综述适合计算机视觉或图形学读者,却不一定回答机器人最关心的问题:这个 3D asset 能不能被仿真器加载?能不能碰撞?能不能开关?能不能产生训练数据?
已有 scene generation surveys 也很多,但它们常关注室内布局、视觉一致性或图形质量。机器人需要的 scene 不只是“沙发在客厅里”,还要支持导航、操作、任务目标、可达性和物理约束。
还有 embodied AI surveys 通常把 3D assets 当成现成基础设施。它们讨论 policy、VLA、simulator、benchmark,却不深入讨论这些 assets 从哪里来、质量怎么评估、怎样变成 URDF/MJCF、怎样进入 sim-to-real pipeline。
因此,旧综述之间有断层:3D 生成懂外观,机器人综述懂任务,仿真综述懂平台,但很少有一篇把“生成内容如何变成机器人可用世界”作为主问题。
所以这一节是想说:本文补的是 3D generation 和 embodied simulation 之间的接口。
这篇论文的新想法
第一,新想法是三角色 taxonomy。Data Generator 把 3D generation 当作 simulation-ready objects and assets 的生产器;Simulation Environments 把 3D generation 当作 interactive worlds 的构建器;Sim2Real Bridge 把 3D generation 当作真实世界和仿真世界之间的连接器。
第二,新想法是强调 simulation-ready。论文反复提到 URDF、MJCF、physics parameters、kinematic structure、material properties、affordance-related semantics。这些不是视觉指标,而是仿真和机器人交互指标。
第三,新想法是指出 field 正在从 visual realism 转向 interaction readiness。外观真实仍然重要,但具身系统更需要 generated content 能被机器人使用。
第四,新想法是把 open challenges 归纳为 limited physical annotations、gap between geometric quality and physical validity、fragmented evaluation、persistent sim-to-real divide 等。它不是只罗列方法,而是指出为什么 3D generation 还不能稳定成为 embodied intelligence 的基础设施。
三角色 taxonomy
┌──────────────────────┐
│ Data Generator │
│ 生成可仿真的物体资产 │
└──────────┬───────────┘
▼
┌──────────────────────┐
│ Simulation Environment│
│ 生成可执行任务的世界 │
└──────────┬───────────┘
▼
┌──────────────────────┐
│ Sim2Real Bridge │
│ 连接真实数据和仿真训练 │
└──────────────────────┘
所以这一节是想说:本文用三角色框架重新组织 3D generation for embodied AI。
它分几步做的(方法)
第 1 步:先定义 simulation-ready 需要什么
输入是一个 3D object 或 scene。普通 3D generation 可能只输出 mesh、point cloud、radiance field 或 texture。具身系统还需要更多东西:kinematic structure、collision mesh、mass、friction、joint limits、material properties、affordance semantics。
处理过程是把外观资产转换成仿真器能理解的结构。比如 URDF 可以描述树状机器人或物体关节结构,MJCF 可以描述 MuJoCo 的 bodies、joints、tendons、actuators、contact parameters。输出是 simulator-compatible asset。
这一步的难点是外观和物理不自动一致。一个柜子看起来像柜子,不代表门能按正确轴旋转;一个杯子 mesh 很漂亮,不代表碰撞体稳定;一块布视觉真实,不代表接触和变形物理正确。
第 2 步:Data Generator 生成可交互资产
Data Generator 角色关注 object-level asset creation。它包括 articulated objects、physically grounded objects、deformable objects 和 end-to-end simulation-ready pipelines。
输入可以是 text、image、mesh、point cloud 或 demonstration。处理可以用 diffusion、LLM/VLM、3DGS、structured latent、procedural rules 等方法。输出要尽量变成可被物理引擎使用的 asset,比如带 articulation 的 URDF、带 material 的 MJCF、带 collision 的 mesh。
这一类方法的核心问题是“物体能不能被机器人操作”。Articulation 让门、抽屉、剪刀有运动结构;physical grounding 让物体有质量和摩擦;deformable modeling 让布料、软体和组织能变形;end-to-end pipeline 让用户从图像或文本直接得到可仿真资产。
第 3 步:Simulation Environments 生成任务世界
Simulation Environments 角色关注 scene-level environment synthesis。它不只生成单个物体,而是生成机器人能进入、导航、操作、完成任务的世界。
输入可能是 task description、layout graph、object list、language instruction 或 agent goal。处理可以是 structure-aware scene synthesis、controllable generation、agentic environment generation。输出是一个可加载到 simulator 的交互场景。
这里的关键是 task-oriented。一个漂亮房间不一定适合训练机器人。如果杯子永远放在不可达位置,抽屉无法打开,导航路径被穿模家具堵住,场景对 robot learning 就没有用。
第 4 步:Sim2Real Bridge 连接真实和仿真
Sim2Real Bridge 角色关注 real-to-sim、sim-to-real 和 closed-loop data lifecycle。它把真实世界观察转换成 digital twin,或者用仿真生成数据增强,再让 learned policy 迁移回真实世界。
输入可能是 RGB-D、video、point cloud、action trajectory、force data 或真实机器人 demonstration。处理是 reconstruction、digital twin construction、domain randomization、data augmentation、synthetic demonstrations、world model simulation。输出是更接近真实部署的训练和评估环境。
这一步解决的是“仿真里学到的东西能不能到真实世界用”。如果 generated assets 只在仿真里好看,但真实机器人一碰就失败,sim-to-real bridge 就断了。
第 5 步:整理 datasets 和 evaluation
Survey 还讨论 datasets and evaluation protocols。对 3D generation 来说,传统指标可能是 FID、Chamfer distance、CLIP score、visual quality;对 embodied AI 来说,还需要 simulator compatibility、task success、physical validity、interaction readiness、sim-to-real success rate。
输入是生成资产和任务评测。处理是用几何、物理、语义和任务指标综合评估。输出是更贴近机器人使用的质量判断。
论文指出 evaluation fragmented 是瓶颈。不同论文用不同 simulator、不同任务、不同资产格式,很难横向比较。
第 6 步:总结挑战和未来方向
最后,论文把挑战归纳为几个方向:physical annotation 不足、geometry quality 与 physical validity 的 gap、deformable/dynamic assets 支持不足、evaluation standards fragmented、sim-to-real divide 持续存在。
这些挑战说明,未来的 3D generation for embodied AI 可能会走向统一 generation-simulation foundation,把生成、仿真、任务评估和真实反馈放进闭环。
所以这一节是想说:本文方法不是提出单一模型,而是建立“生成内容如何进入机器人仿真”的系统框架。

关键数字
| 数字或概念 | 原文语境 | 这说明什么 |
|---|---|---|
| 3 roles | Data Generator / Simulation Environments / Sim2Real Bridge | survey 的主分类框架 |
| 2023 to 2026 | 方法表覆盖近期快速发展阶段 | 领域正处在快速扩张期 |
| URDF / MJCF | simulator-compatible formats | 具身 3D 生成必须输出可执行结构 |
| 4 requirements | geometry、semantics、physics、simulator compatibility 等要求 | 外观之外还有物理和任务约束 |
| 4 bottlenecks | physical annotations、geometry-physics gap、fragmented evaluation、sim-to-real divide | 当前最主要挑战 |
| 3DGen4Robot project page | 作者提供项目页 | 便于继续追踪 taxonomy 和论文列表 |
这篇是 survey,不以单一 benchmark 刷分为主。它的数字更多是分类、范围和挑战结构,而不是一个模型的成功率。
所以这一节是想说:本文的证据是 taxonomy 和领域归纳,不是单模型实验。
实验结果说明了什么
作为 survey,本文没有像模型论文那样报告一个新算法的主实验。它的“结果”是结构化综述:通过整理文献,说明 embodied AI 需要的 3D generation 正在从 visual realism 转向 interaction readiness。
Data Generator 部分说明,物体生成已经开始加入 articulation、physical parameters、URDF/MJCF export。Simulation Environments 部分说明,场景生成开始变得 task-conditioned、controllable、agentic。Sim2Real Bridge 部分说明,生成技术正在进入 digital twin、data augmentation、synthetic demonstration 和 real-to-sim-to-real loop。
这些整理支持一个判断:未来机器人训练世界不是人工手写全部资产,也不是只靠固定 simulator 数据集,而是会越来越多地由 generative models 生成、检查、修正和闭环使用。
但 survey 也提醒我们,现在的 bottleneck 很实在。很多生成结果视觉上可用,但物理上不可用;很多方法能生成单个 object,但不能稳定生成可交互场景;很多评估指标能测几何相似,却不能测真实任务成功。
所以这一节是想说:survey 的结论是方向性证据,告诉我们生成世界要变成机器人基础设施还差哪些环节。
术语表
- Simulation-ready:可以被仿真器加载、碰撞、运动和交互的资产状态。
- Data Generator:把 3D generation 当作可仿真资产生产器。
- Simulation Environments:把生成模型用于构建任务场景和交互世界。
- Sim2Real Bridge:用生成和重建连接真实数据、仿真训练和真实部署。
- URDF:Unified Robot Description Format,常用于描述机器人或关节物体结构。
- MJCF:MuJoCo XML Format,MuJoCo 使用的物理仿真描述格式。
- Digital twin:真实物体或场景的仿真副本。
- Interaction readiness:生成内容能否支持机器人真实交互,而不只是看起来真实。
所以这一节是想说:这篇 survey 的术语都围绕“生成物能不能进仿真器并服务机器人”。
局限和边界
第一,survey 本身不提供一个可直接使用的新 3D generator。它整理领域,而不是发布模型。
第二,taxonomy 很有帮助,但不同方法可能跨多个角色。例如 digital twin 既是 Sim2Real Bridge,也可能是 Data Generator 的输入。
第三,simulation-ready 的评价还没有统一标准。URDF/MJCF 可导出不等于物理准确,task success 也受 controller 和 policy 影响。
第四,deformable 和 dynamic assets 仍然困难。布料、软体、液体、组织等对机器人很重要,但生成和仿真都更难。
第五,sim-to-real divide 仍然存在。生成世界越复杂,越需要真实反馈校准,否则仿真偏差会被放大。
所以这一节是想说:本文建立了框架,但真正的统一生成仿真基础设施还没完成。
和其他论文的关系
和 mimo-embodied 相比,3D generation survey 关注训练世界,而 MiMo 关注跨具身推理模型。一个造环境,一个训练脑。
和 open-h-embodiment 相比,Open-H 汇聚真实医疗机器人数据,本文讨论如何生成仿真资产和环境。真实数据和生成仿真可以互补:真实数据校准仿真,仿真扩展训练覆盖。
和 alanavlm 相比,AlanaVLM 理解第一视角视频,本文讨论如何生成可交互世界。未来可以把生成场景用于训练 first-person embodied QA。
和 habitat、maniskill、isaac 等 simulator 笔记相比,本文更上游。它问的是 simulator 里的 object、scene、digital twin 从哪里来。
所以这一节是想说:本文连接了 simulation、dataset、world model 和 robot policy 四条线。
和本导读的关系
本站的 simulation 主题以前多关注 Habitat、MuJoCo、Isaac Gym、SIMPLER、ManiSkill 等平台。3D Generation Survey 提醒我们,平台之外还有一个更上游的问题:训练世界的内容如何规模化产生。
它适合放在 sim 主题,也适合和 world-model、dataset-eval、VLA 主题一起读。读者可以用它理解为什么“更多仿真”不只是开更多环境,还要生成可交互、可评估、能迁移的 3D 内容。
所以这一节是想说:这篇 survey 补齐本站从仿真平台到生成式仿真内容的链路。
思考题
- 为什么外观逼真的 3D object 不一定适合机器人训练?
- URDF 和 MJCF 在 embodied 3D generation 中为什么重要?
- Data Generator、Simulation Environments、Sim2Real Bridge 三个角色有什么区别?
- 为什么 geometry quality 和 physical validity 之间会有 gap?
- 如果你要训练开柜门机器人,生成世界至少需要哪些非视觉信息?
FAQ
Q:这篇论文是不是提出了新的 3D 生成模型? A:不是。它是一篇 survey,主要贡献是 taxonomy、文献整理和挑战归纳。
Q:simulation-ready 是不是只要导出 URDF 就够? A:不够。URDF/MJCF 是格式,物理参数、碰撞体、关节、材料和任务可用性都要验证。
Q:为什么 3D generation 和 sim-to-real 有关系? A:生成技术可以创建 digital twin、增强数据、生成 demonstrations,但这些仿真内容必须和真实世界对齐,才能帮助迁移。
Q:本站有没有跑 3D generation? A:没有。这里只记录 survey 的分类和观点,不写成本地复现实验。
进一步读什么
- Habitat / AI2-THOR / ManiSkill / MuJoCo:理解生成资产最终要进入什么仿真平台。
- URDF-Anything / URDFormer:理解从视觉到可动关节描述的路线。
- RoboTwin / RoboCasa:理解生成式仿真环境如何服务 robot learning。
open-h-embodiment:理解真实数据如何补充仿真和生成。
精读补充:为什么 simulation-ready 比 visual realism 更难
这篇 survey 最值得反复记住的转向,是从 visual realism 到 interaction readiness。视觉逼真主要服务人眼判断,simulation-ready 则服务机器人训练。人看到一个柜子,会默认门能打开;仿真器不会默认知道这件事。它需要 joint axis、joint limit、collision mesh、mass、friction、contact model 等显式信息。
这种差异解释了为什么 3D generation for embodied AI 不能只沿用图形学指标。一个模型的 Chamfer distance 很低,说明几何接近;纹理很漂亮,说明渲染质量高。但如果抽屉没有正确滑轨,机器人 policy 在仿真里学到的开抽屉动作就会错误。几何质量和物理有效性之间的 gap,正是论文反复强调的 bottleneck。
URDF 和 MJCF 也不只是文件格式。它们代表生成结果能否进入机器人软件栈。URDF 更常见于机器人结构描述,MJCF 更贴近 MuJoCo 物理仿真。一个生成 pipeline 如果能输出这些结构,就从“给人看的模型”更接近“给机器人训练用的资产”。但格式正确仍不等于物理正确,参数还需要校准和验证。
未来更理想的系统可能是闭环的:从真实世界扫描或语言任务生成初始 3D asset,放进 simulator 跑机器人交互,发现碰撞、关节或材料不合理,再把失败反馈回生成器修正。只有这种 generate-simulate-validate loop 成熟后,3D generation 才可能真正成为 robot foundation model 的训练世界工厂。
所以这一节是想说:simulation-ready 难在物理、格式、任务和验证闭环,而不只是模型外观。
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:三角色 taxonomy 是否为 Data Generator、Simulation Environments、Sim2Real Bridge;simulation-ready 相关要求是否包含 kinematics、physics、URDF/MJCF 等;四类 bottleneck 是否按原文表述;project page 是否为 https://3dgen4robot.github.io;不要把 survey 观点写成本站实验结论。
原文信息
- arXiv: 2604.26509
- PDF: https://arxiv.org/pdf/2604.26509
- Project: https://3dgen4robot.github.io
@article{ye2026generationembodied,
title = {3D Generation for Embodied AI and Robotic Simulation: A Survey},
author = {Ye, Tianwei and Mao, Yifan and Liao, Minwen and Liu, Jian and Guo, Chunchao and Du, Dazhao and Shou, Quanxin and Zhu, Fangqi and Guo, Song},
journal = {arXiv preprint arXiv:2604.26509},
year = {2026}
}
◼
引用本笔记 / Cite this note
@online{eai_embodied_3d_generation_survey_2026,
title = {(readable note) 3D Generation for Embodied AI and Robotic Simulation: A Survey},
author = {Xun, Jason},
year = {2026},
note = {Note on a 2026 paper},
howpublished = {\url{https://estelledc.github.io/embodied-ai-reading-station/papers/embodied-3d-generation-survey/}},
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?