RoboMIND2.0

Name: RoboMIND2.0
Creator: maas
Published: 2026-07-12 01:22:40
License: 暂无描述

魔搭社区2026-07-12 更新2026-01-10 收录

下载链接：

https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0

下载链接

链接失效反馈

官方服务：

资源简介：

<a href="https://opensource.x-humanoid-cloud.com/plugin.php?id=keke_video_base&ac=course&cid=19"> <img src="./assets/robomind-v2-dataset.png"> </a> ![RoboMIND2.0](assets/teaser.jpg) # 🎉🎉 RoboMIND2.0 🎉🎉 [中文](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0/file/view/master/README.zh.md?id=169304&status=1) | [English](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0/file/view/master/README.md?id=169304&status=1) RoboMIND2.0 has arrived. This release brings a major upgrade over RoboMIND. We focus on complex data from **dual-arm manipulation** and **mobile manipulation** scenarios, which are highly demanded by the community. RoboMIND2.0 introduces many practical features to further advance robot learning and embodied intelligence research. Wanna review RoboMIND1.0? Please check modelscope repo [RoboMIND MS](https://modelscope.cn/datasets/X-Humanoid/RoboMIND) or Huggingface repo [RoboMIND HF](https://huggingface.co/datasets/x-humanoid-robomind/RoboMIND). ## Key Upgrades - RoboMIND2.0 scales up to **310K trajectories**, totaling over **1,000 hours** of data. - RoboMIND2.0 includes **12K tactile-enriched sequences**. - RoboMIND2.0 collects **20K mobile manipulation trajectories**. - RoboMIND2.0 covers **6 popular robotic embodiments**. ## Embodiments Involved 🤖 <table> <tr> <td align="center"><img src="assets/tienkung.jpg" width="200" alt="Tienkung"></td> <td align="center"><img src="assets/tienyi.jpg" width="200" alt="Tianyi"></td> <td align="center"><img src="assets/agilex.jpg" width="200" alt="AgileX"></td> </tr> <tr> <td align="center">Tienkung</td> <td align="center">Tianyi</td> <td align="center">AgileX</td> </tr> <tr> <td align="center"><img src="assets/franka.jpg" width="200" alt="Franka"></td> <td align="center"><img src="assets/ur.jpg" width="200" alt="UR"></td> <td align="center"><img src="assets/ark.jpg" width="200" alt="Ark"></td> </tr> <tr> <td align="center">Franka</td> <td align="center">UR</td> <td align="center">Ark</td> </tr> </table> ### Sub-repository Links - **Tienkung** - [X-Humanoid/RoboMIND2.0-Tienkung](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Tienkung) - **Tianyi** - [X-Humanoid/RoboMIND2.0-Tianyi](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Tianyi) - **Tianyi (Mobile)** - [X-Humanoid/RoboMIND2.0-Tianyi-mobile](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Tianyi-mobile) - **AgileX** - [X-Humanoid/RoboMIND2.0-Agilex](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Agilex) - **AgileX (Mobile)** - [X-Humanoid/RoboMIND2.0-Agilex-mobile](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Agilex-mobile) - **Franka** - [X-Humanoid/RoboMIND2.0-Franka-Part-1](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Franka-Part-1) - [X-Humanoid/RoboMIND2.0-Franka-Part-2](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Franka-Part-2) - [X-Humanoid/RoboMIND2.0-Franka-Part-3](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Franka-Part-3) - [X-Humanoid/RoboMIND2.0-Franka-Part-4](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Franka-Part-4) - [X-Humanoid/RoboMIND2.0-Franka-Part-5](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Franka-Part-5) - **UR** - [X-Humanoid/RoboMIND2.0-UR5](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-UR5) - **UR + DexHand** - [X-Humanoid/RoboMIND2.0-UR5-Dex](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-UR5-Dex) - **Ark** - [X-Humanoid/RoboMIND2.0-Ark](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Ark) - **Ark (Mobile)** - [X-Humanoid/RoboMIND2.0-Ark-mobile](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Ark-mobile) - **Simulation** - [X-Humanoid/RoboMIND2.0-Tienkung-sim](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Tienkung-sim) - [X-Humanoid/RoboMIND2.0-Franka-sim](https://modelscope.cn/datasets/X-Humanoid/RoboMIND2.0-Franka-sim) ## Overall HDF5 Data Structure Below is the unified HDF5 format used in RoboMIND2.0. Each dataset variant contains a subset of the following structure. ``` / ├── metadata │ ├── language_instruction │ ├── data_type │ ├── data_format_version │ ├── collection_time │ ├── collector │ ├── sim_assets │ └── trajectory_length │ ├── camera_model │ ├── camera_top │ ├── camera_front │ ├── camera_left │ ├── camera_right │ ├── camera_head │ ├── camera_wrist_left │ └── camera_wrist_right │ ├── camera_color_resolution │ ├── camera_top │ ├── camera_front │ ├── camera_left │ ├── camera_right │ ├── camera_head │ ├── camera_wrist_left │ └── camera_wrist_right │ ├── camera_color_channel │ ├── camera_top │ ├── camera_front │ ├── camera_left │ ├── camera_right │ ├── camera_head │ ├── camera_wrist_left │ └── camera_wrist_right │ ├── camera_depth_resolution │ ├── camera_top │ ├── camera_front │ ├── camera_left │ ├── camera_right │ ├── camera_head │ ├── camera_wrist_left │ └── camera_wrist_right │ ├── camera_intrinsics │ ├── camera_top │ │ ├── matrix │ │ └── dist_coeffs │ ├── camera_front │ │ ├── matrix │ │ └── dist_coeffs │ ├── camera_left │ │ ├── matrix │ │ └── dist_coeffs │ ├── camera_right │ │ ├── matrix │ │ └── dist_coeffs │ ├── camera_head │ │ ├── matrix │ │ └── dist_coeffs │ ├── camera_wrist_left │ │ ├── matrix │ │ └── dist_coeffs │ └── camera_wrist_right │ ├── matrix │ └── dist_coeffs │ ├── camera_extrinsics │ ├── camera_top │ ├── camera_front │ ├── camera_left │ ├── camera_right │ ├── camera_head │ ├── camera_wrist_left │ └── camera_wrist_right │ ├── base_to_robot_transformation │ ├── base_to_robot_left │ ├── base_to_robot_right │ └── base_to_robot_single │ ├── force_model │ ├── force_left_end_effector │ ├── force_right_end_effector │ ├── force_left_arm_joints │ └── force_right_arm_joints │ ├── tactile_model │ ├── tactile_left │ └── tactile_right │ ├── master │ ├── arm_left_position_raw │ │ ├── timestamp │ │ ├── is_intervene │ │ └── data │ ├── arm_left_position_align │ ├── arm_right_position_raw │ ├── arm_right_position_align │ ├── arm_single_position_raw │ ├── arm_single_position_align │ ├── end_effector_left_position_raw │ ├── end_effector_left_position_align │ ├── end_effector_right_position_raw │ ├── end_effector_right_position_align │ ├── end_effector_single_position_raw │ ├── end_effector_single_position_align │ ├── end_effector_left_pose_raw │ ├── end_effector_left_pose_align │ ├── end_effector_right_pose_raw │ ├── end_effector_right_pose_align │ ├── end_effector_single_pose_raw │ ├── end_effector_single_pose_align │ ├── head_position_raw │ ├── head_position_align │ ├── waist_position_raw │ ├── waist_position_align │ ├── leg_left_position_raw │ ├── leg_left_position_align │ ├── leg_right_position_raw │ ├── leg_right_position_align │ ├── chassis_pose_raw │ ├── chassis_pose_align │ ├── chassis_twist_raw │ └── chassis_twist_align │ ├── puppet │ └── (same structure as master) │ ├── camera_observations │ ├── timestamp │ ├── is_intervene │ ├── color_images │ │ ├── camera_top │ │ ├── camera_front │ │ ├── camera_left │ │ ├── camera_right │ │ ├── camera_head │ │ ├── camera_wrist_left │ │ └── camera_wrist_right │ └── depth_images │ ├── camera_top │ ├── camera_front │ ├── camera_left │ ├── camera_right │ ├── camera_head │ ├── camera_wrist_left │ └── camera_wrist_right │ ├── force_observations │ ├── force_left_end_effector_raw │ ├── force_left_end_effector_align │ ├── force_right_end_effector_raw │ ├── force_right_end_effector_align │ ├── force_left_arm_joints_raw │ ├── force_left_arm_joints_align │ ├── force_right_arm_joints_raw │ └── force_right_arm_joints_align │ └── tactile_observations ├── tactile_left_raw ├── tactile_left_align ├── tactile_right_raw └── tactile_right_align ``` ## RoboMIND-Sim Additionally, we have open-sourced a simulation environment built on Isaac Sim, which includes training datasets and standardized evaluation scripts. Researchers can leverage these resources out-of-the-box to rapidly conduct model performance evaluation and algorithm validation within the simulation. [RoboMIND-Sim](https://github.com/Open-X-Humanoid/RoboMIND-Sim) ## Discussion: <td align="left"><img src="assets/wechat.jpg" width="400" alt="Tienkung"></td> ## Citation If you find this project helpful, please cite: ``` @misc{hou2025robomind20multimodalbimanual, title={RoboMIND 2.0: A Multimodal, Bimanual Mobile Manipulation Dataset for Generalizable Embodied Intelligence}, author={Chengkai Hou and Kun Wu and Jiaming Liu and Zhengping Che and Di Wu and Fei Liao and Guangrun Li and Jingyang He and Qiuxuan Feng and Zhao Jin and Chenyang Gu and Zhuoyang Liu and Nuowei Han and Xiangju Mi and Yaoxu Lv and Yankai Fu and Gaole Dai and Langzhe Gu and Tao Li and Yuheng Zhang and Yixue Zhang and Xinhua Wang and Shichao Fan and Meng Li and Zhen Zhao and Ning Liu and Zhiyuan Xu and Pei Ren and Junjie Ji and Haonan Liu and Kuan Cheng and Shanghang Zhang and Jian Tang}, year={2025}, eprint={2512.24653}, archivePrefix={arXiv}, url={https://arxiv.org/abs/2512.24653}, } ```

RoboMIND2.0 is a large-scale multimodal robotic manipulation dataset developed for embodied intelligence and robot learning, with a particular emphasis on complex bimanual manipulation and mobile manipulation scenarios. As a major extension of RoboMIND 1.0, the dataset contains approximately 310,000 robot trajectories totaling more than 1,000 hours, including around 12,000 tactile-enriched sequences and 20,000 mobile manipulation trajectories collected across six widely used robotic platforms: Tienkung, Tianyi, AgileX, Franka, UR, and Ark. RoboMIND2.0 adopts a unified HDF5 data structure that integrates multi-view RGB-D observations, robot joint and end-effector states, control signals, chassis motion information, force and tactile feedback, language instructions, camera calibration parameters, and simulation assets. It supports research on robot policy learning, vision-language-action model training, cross-embodiment transfer, coordinated dual-arm manipulation, and mobile manipulation. The project also provides RoboMIND-Sim, an NVIDIA Isaac Sim-based simulation environment with training data and standardized evaluation scripts, facilitating algorithm validation, model benchmarking, and the development of generalizable embodied intelligence.

提供机构：

maas

创建时间：

2026-01-04

搜集汇总

数据集介绍