When high-quality action data transforms from “private assets” to “public infrastructure”, Embodied AI is expected to shift from single-point breakthroughs to the systematic emergence. The scaled supply of data will significantly lower the threshold for entrepreneurship, accelerate algorithm iteration, and give rise to the blooming of various vertical scenarios.
On June 16, 2026, as a global leader in motion capture, CHINGMU, in collaboration with EnDecodeX, launched the “MotionDecode Data Openness Program”, offering free applications for the first phase of 1000-hour of high-quality Embodied Motion Dataset worldwide, injecting strong data momentum into the training of Embodied AI model and robot motion learning.
01.MotionDecode Data Openness Program
At the critical stage when embodied intelligence is accelerating towards industrial application, “data-driven” has become a consensus in the industry. How to enable robots to understand human movements and adapt to the real world is the primary prerequisite for achieving technological leaps.
“MotionDecode” is a multimodal high-precision motion data system constructed by CHINGMU for robot enterprises, embodied AI algorithm teams, universities and research institutions, simulation platforms and model teams. The first phase of this plan focuses on core scenarios such as robot training, motion generation, simulation verification, and learning of embodied AI models , opening 1000-hour high-quality human motion data resources for free, promoting the industry to move from single-point technology exploration to the joint construction of a data ecosystem.
Core Features
● Larger Scale – A 3,000-hour data reserve has been established. The first phase of 1,000 hours of high-precision human movement data is freely available, with an expected annual output of 500,000 hours.
● Higher precision – Based on the CHINGMU optical motion capture system for data collection,providing sub-millimeter precision ability, with joint angle error less than 0.5°, ensuring industrial-grade accuracy of data and meeting the strict requirements of robot control training.
●More Modalities – Multimodal data such as Body, Hand (including force feedback), Ego,Exo, object 6D, environmental scene, etc. The scale of each modal data is as follows:
(a) Full – body motion data：200+ hours, covering 100+ action types (walking, running, jumping, carrying, etc.)
(b) Object position & rigid – body tracking：150+ hours, with 6DoF synchronous capture of hands, props, and objects in the same coordinate system (100+ rigid bodies)
(c) Robot retargeting data：300+ hours, featuring human – to – robot skeleton retargeting, simulation validation, and quality reports, adaptable to multiple robot models
(d) Multi – view synchronized video：400+ hours, with 8 – camera sync and frame – alignment to motion capture, ready for vision model input and validation
● Broader Scene – Covering typical application scenarios such as industrial manufacturing, home services, supermarket retail, medical rehabilitation, logistics and warehousing, ball game interaction, and digital entertainment, it includes 500+ human-machine tasks and 200+ types of object tools and props, providing abundant materials for algorithm training in vertical scenarios.
● Easier Application – Compatible with over 10 common data formats such as BVH, FBX, and CSV, supporting adaptation to mainstream robot simulation formats like MuJoCo/Isaac/URDF, lowering the access threshold and allowing algorithm teams to focus on the model itself.
Data Production Process
The stability of data quality determines the upper limit of model training. From scene design to dataset release, MotionDecode has established an 8-step standardized production process, each of which is auditable and reproducible, ensuring stable and controllable data quality
Scenario Design → Optical Motion Capture → Finger & Object Tracking → Multi-view Video Sync → Cleaning & Annotation → Robot Retargeting → Quality Sssessment → Dataset Release
02 Multi-scene Action Demonstration
MotionDecode dataset covers a rich variety of action types. After robot redirection, they can be directly transferred to humanoid robots. The following are some representative action demonstrations:
(OR Video: https://youtu.be/siRGnneV_v0?si=jJDAs5-mo6-qkvfg)
03 Acquisition Methods
We sincerely invite you to join the MotionDecode data early trial and feedback program. The first phase of the 1000-hour high-quality human motion dataset has opened a free application channel by scanning the code and will continue to be updated and iterated. More example data will be gradually made available through GitHub and Hugging Face. Please pay attention to the latest information on the CHINGMU Official Website.
Project Homepage: https://chingmudata.github.io/MotionDecode/
Dataset: https://huggingface.co/datasets/CMRobot/MotionDecode
Website: https://en.chingmu.com/
The true barrier of Robotics does not lie in data monopoly, but the continuous breakthroughs in understanding depth and generalization of physical world. The MotionDecode Data Openness Program will significantly lower the R&D threshold for the embodied AI industry, providing a unified and high-quality training data benchmark for global research institutions and enterprises, and filling the gap in the industry in the field of large-scale, high-precision, and all-scenario human motion data to empower breakthroughs in full-body motion control of humanoid robots, precise operation of dexterous hands, and human-robot collaborative interaction technologies, accelerate the rapid transition of embodied AI from the technical verification stage to large-scale industrial application.
In the future, CHINGMU will continue to deepen technological R&D and data capacity construction, join hands with partners from the industrial chain to jointly improve the standard and ecosystem of embodied AI, promoting embodied AI from technological exploration to large-scale application, and empowering the intelligent upgrade of industries.
Media ContactCompany Name: SHANGHAI CHINGMU VISION TECHNOLOGY CO., LTDContact Person: CHINGMUEmail: Send EmailPhone: 13310099680City: ShanghaiCountry: ChinaWebsite: https://en.chingmu.com/
