# Embodied AI Evidence Hub / 具身智能证据知识库

> Evidence-first Chinese research hub for embodied AI, robot learning, VLA, world models, multimodal sensing, 4D reasoning, data quality, and evaluation.

## Research method

Each public topic is derived from a settled evidence run. Accepted evidence requires complete non-OCR full text, a validated paper note, and a passing claim-support audit. Static pages expose the full Zhihu explainer, deduplicated paper citations, evidence metrics, and an auditable appendix.

## Canonical directory

- [Homepage](https://yuanzh0u.github.io/embodied-learning/)
- [Research directory](https://yuanzh0u.github.io/embodied-learning/research/)
- [智驾模型-vla-驾驶基础模型与关键团队-理想增补 / Autonomous-Driving VLA, Foundation Models, and Key Teams](https://yuanzh0u.github.io/embodied-learning/research/autonomous-driving-vla-models-teams/)
- [无本体ego-centric数据中手部检测精度对本体末端轨迹的影响 / Hand-Detection Precision vs. End-Effector Trajectory in Egocentric Data](https://yuanzh0u.github.io/embodied-learning/research/egocentric-hand-precision-vs-trajectory/)
- [近一年鱼眼镜头在具身领域的应用 / Fish-Eye Cameras in Embodied AI](https://yuanzh0u.github.io/embodied-learning/research/fisheye-cameras-in-embodied-ai/)
- [近一年机器人末端轨迹精度的解决方案与痛点 / End-Effector Trajectory Accuracy: Solutions and Pain Points](https://yuanzh0u.github.io/embodied-learning/research/end-effector-trajectory-accuracy/)
- [近一年SLAM技术在具身智能领域是否有核心作用 / Is SLAM Core to Embodied Intelligence?](https://yuanzh0u.github.io/embodied-learning/research/slam-in-embodied-ai/)
- [近一年类脑模型的发展 / Brain-Inspired Models over the Past Year](https://yuanzh0u.github.io/embodied-learning/research/brain-inspired-models/)
- [Ego-Exo 相机配置与视角对齐：第三人称相机如何初始化/标定、第一↔第三人称视角如何对齐、多台第三人称相机如何对齐 / Ego–Exo Camera Calibration and View Alignment](https://yuanzh0u.github.io/embodied-learning/research/ego-exo-camera-calibration-alignment/)
- [Ego-Exo 后继研究：第三人称→第一人称视觉表征迁移的演进 / Ego–Exo Representation Transfer](https://yuanzh0u.github.io/embodied-learning/research/ego-exo-representation-transfer/)
- [具身端到端模型除动作监督外还需要哪些监督信号 / Supervision Beyond Actions for End-to-End Embodied Models](https://yuanzh0u.github.io/embodied-learning/research/embodied-model-supervision-beyond-actions/)
- [世界模型训练是否有必要接监督信号还是走纯端到端 / World Model Supervision vs. End-to-End Learning](https://yuanzh0u.github.io/embodied-learning/research/world-model-supervision-vs-end-to-end/)
- [ACT及动作分块策略在RoboTwin 2.0中的接入、训练与闭环评测 / ACT and Action Chunking in RoboTwin 2.0](https://yuanzh0u.github.io/embodied-learning/research/act-robotwin-action-chunking/)
- [近一年 Ego-centric 数据手部检测与追踪的问题和难点 / Challenges in Egocentric Hand Tracking](https://yuanzh0u.github.io/embodied-learning/research/egocentric-hand-tracking-challenges/)
- [近一年空间数据生产难点及具身机器人与智能驾驶数据难点异同 / Spatial Data Production for Robotics and Autonomous Driving](https://yuanzh0u.github.io/embodied-learning/research/spatial-data-production-robotics-vs-autonomous-driving/)
- [近一年具身智能论文中的数据时空一致性 / Spatiotemporal Consistency in Embodied AI Data](https://yuanzh0u.github.io/embodied-learning/research/embodied-data-spatiotemporal-consistency/)
- [近一年具身智能预训练模型对数据源与采集参数的要求 / Data Requirements for Embodied Pretraining](https://yuanzh0u.github.io/embodied-learning/research/embodied-pretraining-data-requirements/)
- [近半年智能体技术在具身智能行业的发展应用 / Agentic AI Applications in Embodied Intelligence](https://yuanzh0u.github.io/embodied-learning/research/embodied-ai-agent-industry-applications/)
- [近三年具身机器人原子技能的发展及VLA成为主流的技术原因 / Robot Atomic Skills and the Rise of VLA](https://yuanzh0u.github.io/embodied-learning/research/robot-atomic-skills-and-vla/)
- [LeWorldModel 技术谱系：JEPA 潜空间世界模型到规划控制 / LeWorldModel: From JEPA to Planning and Control](https://yuanzh0u.github.io/embodied-learning/research/leworldmodel-jepa-planning-control/)
- [近一年触觉数据与视觉数据联合训练的方法和进展 / Visual–Tactile Joint Training](https://yuanzh0u.github.io/embodied-learning/research/visual-tactile-joint-training/)
- [近半年力觉在具身机器人领域的发展 / Progress in Force Sensing for Robotics](https://yuanzh0u.github.io/embodied-learning/research/force-sensing-robotics-progress/)
- [近一年触觉在具身机器人领域的发展 / Progress in Tactile Sensing for Robotics](https://yuanzh0u.github.io/embodied-learning/research/tactile-robotics-progress/)
- [近一年触觉、力觉、视觉、语言等多模态数据在具身机器人训练方法中的演进 / Multimodal Training for Embodied Robots](https://yuanzh0u.github.io/embodied-learning/research/multimodal-embodied-robot-training/)
- [近半年VLA在具身领域最大的技术突破 / Recent Breakthroughs in VLA Models](https://yuanzh0u.github.io/embodied-learning/research/vla-breakthroughs/)
- [近一年世界视频模型最可靠的应用任务 / Reliable Tasks for Video World Models](https://yuanzh0u.github.io/embodied-learning/research/reliable-video-world-model-tasks/)
- [近一年 loco-manipulation 研究进展 / Progress in Loco-Manipulation](https://yuanzh0u.github.io/embodied-learning/research/loco-manipulation-progress/)
- [近一年为何说反应式VLA已死世界模型当立 / Reactive VLA vs. World Models](https://yuanzh0u.github.io/embodied-learning/research/reactive-vla-vs-world-models/)
- [近一年论文中的具身数据污染问题 / Data Contamination in Embodied AI](https://yuanzh0u.github.io/embodied-learning/research/embodied-data-contamination/)
- [近一年图像视觉定位方法的发展与挑战 / Progress and Challenges in Visual Localization](https://yuanzh0u.github.io/embodied-learning/research/visual-localization-progress/)
- [Ego-centric 数据在具身模型训练中的问题与困难 / Egocentric Data for Embodied Model Training](https://yuanzh0u.github.io/embodied-learning/research/egocentric-data-for-robot-learning/)
- [近半年 UMI 数据质量 / UMI Data Quality](https://yuanzh0u.github.io/embodied-learning/research/umi-data-quality/)
- [近一年已发表论文中的具身智能数据质量 / Data Quality in Embodied AI](https://yuanzh0u.github.io/embodied-learning/research/embodied-ai-data-quality/)
- [近一年具身感知问题是否已有效解决 / Has Embodied Perception Been Solved?](https://yuanzh0u.github.io/embodied-learning/research/embodied-perception-solved/)
- [近一年具身感知与导航是否已解决 / Have Embodied Perception and Navigation Been Solved?](https://yuanzh0u.github.io/embodied-learning/research/embodied-perception-navigation-solved/)
- [近一年具身导航问题是否已有效解决 / Has Embodied Navigation Been Solved?](https://yuanzh0u.github.io/embodied-learning/research/embodied-navigation-solved/)
- [触觉世界模型 / Tactile World Models](https://yuanzh0u.github.io/embodied-learning/research/tactile-world-models/)
- [具身智能数据质量的主要矛盾 / Core Contradictions in Embodied AI Data Quality](https://yuanzh0u.github.io/embodied-learning/research/embodied-data-quality-contradictions/)
- [具身数据感知误差溯源 / Tracing Perception Errors in Embodied Data](https://yuanzh0u.github.io/embodied-learning/research/embodied-perception-error-traceability/)
- [具身数据感知误差与认知误差区别 / Perception Errors vs. Cognitive Errors](https://yuanzh0u.github.io/embodied-learning/research/perception-vs-cognitive-errors/)
- [具身传感器感知误差 / Sensor Perception Errors in Embodied AI](https://yuanzh0u.github.io/embodied-learning/research/embodied-sensor-perception-errors/)
- [世界模型需要什么样的训练数据 / What Training Data Do World Models Need?](https://yuanzh0u.github.io/embodied-learning/research/world-model-training-data/)
- [世界模型评测边界 / Evaluation Boundaries for World Models](https://yuanzh0u.github.io/embodied-learning/research/world-model-evaluation-boundaries/)
- [Sparse language, dense vision, and continuous action alignment in VLA systems / Language, Vision, and Action Alignment in VLA Systems](https://yuanzh0u.github.io/embodied-learning/research/vla-language-vision-action-alignment/)
- [4D时空推理对数据的需求 / Data Requirements for 4D Spatiotemporal Reasoning](https://yuanzh0u.github.io/embodied-learning/research/4d-reasoning-data-requirements/)
- [4D时空推理 / 4D Spatiotemporal Reasoning](https://yuanzh0u.github.io/embodied-learning/research/4d-spatiotemporal-reasoning/)

- [Mind-Omni 图解阅读](https://yuanzh0u.github.io/embodied-learning/research/autonomous-driving-vla-models-teams/mind-omni/)

## Reuse and citation

Code and research content have separate licenses. Cite the canonical topic URL and the underlying papers; see the repository CITATION.cff and LICENSES.md for details.