All 具身智能公司的数据从哪来？开源数据与商业实战数据 - Naifan Li's Bloghttps://blog.omagiclee.com/posts/embodied-intelligence/data/All 具身智能公司的数据从哪来？开源数据与商业实战数据 | Naifan Li's BlogHugo -- gohugo.iozh-CNTue, 31 Mar 2026 13:18:51 +0800京东具身数据https://blog.omagiclee.com/posts/embodied-intelligence/data/joyegocam/Tue, 21 Apr 2026 13:33:45 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/joyegocam/ arXiv

【26.04.16京东具身智能生态发布会，发布具身大模型JoyAI-RA、具身智能交易平台】https://www.bilibili.com/video/BV1H8dYBTEax?vd_source=e05fafaa995c59e16affc059ba418214

]]>Universal Manipulation Interface: In-The-Wild Robot Teaching Without In-The-Wild Robotshttps://blog.omagiclee.com/posts/embodied-intelligence/data/umi/Thu, 16 Apr 2026 19:53:52 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/umi/ arXiv 2402.10329 Project

鹿鸣机器人， 遥操，一天采集100-120条数据，人工费200块，一条成本两块钱。

fast umi 一天采集500+，4毛到5毛左右；夹爪成本和机器人本体成本降低，一个本体价格大概在30-40万人民币，

]]>Indexhttps://blog.omagiclee.com/posts/embodied-intelligence/data/aoe/Thu, 16 Apr 2026 14:20:11 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/aoe/ arXiv

Data collection:

  • Hand-held Gripper-based Interfaces
    • UMI
    • FastUMI
    • DexUMI
  • Wearable AR/VR Systems Humanoidexo: Scalable whole-body humanoid manipulation via wearable exoskeleton
  • Passive Egocentric Videos

Always-On Egocentric Data System

Hardware

Mobile Software

]]>Indexhttps://blog.omagiclee.com/posts/embodied-intelligence/data/shanghai-robot-gov/Wed, 15 Apr 2026 11:01:22 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/shanghai-robot-gov/<p>人形机器人具身智能训练场（以下简称“训练场”）由国地中心牵头建设，位于浦东张江模力社区，建设面积超5000平，首期已部署超过100台异构人形机器人，是全国首个异构人形机器人训练场。</p>Indexhttps://blog.omagiclee.com/posts/embodied-intelligence/data/nils/Mon, 13 Apr 2026 16:36:41 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/nils/ arXiv

TL;DR

Identifying Objects in the Scene

-> objects and properties.

VLMs, Grounding DINO, SigLIP

Object-Centric Scene Annotation

using an ensemble of open-vocabulary detection and segmentation models.

Keystate Detection and Label Generation

]]>Robo2VLMhttps://blog.omagiclee.com/posts/embodied-intelligence/data/robo2vlm/Mon, 13 Apr 2026 16:02:27 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/robo2vlm/ arXiv

TL;DR

Motivations & Innovations

Data Recipe

Scene-Interaction Understanding

Embodied Scene Understanding:

Input:

  • a task description in nature language
  • images from different camera views

Output:

  • using an off-the-shelf LLM such as Qwen 2.5 to obtain target object, scene, task and skill description.

Manipulation Phase Segmentation:

]]>DROID: A Large-Scale In-The-Wild Robot Manipulation Datasethttps://blog.omagiclee.com/posts/embodied-intelligence/data/droid/Mon, 13 Apr 2026 15:20:56 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/droid/ arXiv

TL;DR

Motivations & Innovations

Approach

Model

Training Recipe

Data Recipe

Experiments

]]>Open X-Embodiment: Robotic Learning Datasets and RT-X Modelshttps://blog.omagiclee.com/posts/embodied-intelligence/data/oxe/Mon, 13 Apr 2026 14:40:49 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/oxe/ arXiv

Open X-Embodiment Dataset

robot learning dataset with 1M+ robot trajectories from 22 robot embodiments.

Dataset Analysis

Pre-Trained Checkpoints

a selection of RT-X model checkpoints ready for inference and finetuning.

TL;DR

]]>Indexhttps://blog.omagiclee.com/posts/embodied-intelligence/data/summary/Mon, 13 Apr 2026 14:10:49 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/summary/ arXiv

EgoCentric，iPhone 固定在头上，录制自己叠衣服、洗碗、做饭的画面

Micro1 Scale AI、Encord等数据公司都在招募自己的数据采集大军，DoorDash 则付钱给外卖员拍摄做家务的视频。

]]>UniACT-datasethttps://blog.omagiclee.com/posts/embodied-intelligence/data/uniact-dataset/Tue, 07 Apr 2026 14:49:10 +0800https://blog.omagiclee.com/posts/embodied-intelligence/data/uniact-dataset/ arXiv

From six public datasets, we clean, standardize, and balance samples to construct UniACT-dataset, a large-scale dataset with over 6 million trajectories and 9,500 hours of data,

]]>