On August 27, CHINGMU Shanghai Vision Technology Co., Ltd. officially joined the OpenLET community. As a global leader in motion capture, CHINGMU will leverage its decade‑long full‑stack expertise in optical motion capture to deeply participate in the development and construction of human motion data within the community, working together to solidify the"motion data foundation" for the Embodied AI era.

Donating Mocap Data to Co‑Build HumanMotion

The quality of data acquisition directly determines the upper bound of motion policy models.

Upon joining the community, CHINGMU will participate in the co‑development of the SIG 「HumanMotion」 working group, focusing on three core areas: acquisition standards for motion capture data, data formats, and quality assessment. Our goal is to translate the collection workflows that have been repeatedly validated in real‑world projects into reusable technical experience for the community, enabling more teams to produce human motion data that is trainable, reproducible, and comparable at lower cost.

In addition, CHINGMU will make an initial donation of human motion capture data to the community, with motion retargeting and format adaptation completed for the Roban robot. The adapted data can be directly used for cerebellar motion policy training and algorithm validation. Specific details regarding scale, format, and access methods will be announced by the community in due course.

From Collection Capability to Data Ecosystem – Continuously Opening Up the Value of Motion Perception

In the Embodied AI field, CHINGMU has already built the MotionDecode Embodied AI dataset – a multimodal, high‑quality motion data system for robotics companies, algorithm teams, research institutions, and simulation platforms. To date, 3,000 hours of data have been accumulated, with an estimated annual production capacity of 500,000 hours. The dataset covers multimodal data including body pose, hand pose, ego video, tactile pressure, IMU, third‑person view, interactive object pose, and environmental data across diverse real‑world scenarios such as home, industrial, office, and outdoor environments, and is compatible with over ten standard formats including BVH, FBX, and CSV.

In its first phase, the program has already made 1,000 hours of high‑quality human motion data freely available for core scenarios including robot training, motion generation, simulation validation, and Embodied AI model learning – driving the industry from isolated technical exploration toward collaborative data ecosystem building.

We invite you to join the MotionDecode Early Access & Feedback Program. The first 1,000 hours of high‑quality human motion data are now open for free application via QR code, with continuous updates rolling out. More sample data will be released on GitHub, Hugging Face, and the CHINGMU official website – stay tuned.

MotionDecode Data Open Program has significantly lowered the R&D barrier for the Embodied AI industry, providing a unified, high-quality training data benchmark for research institutions and enterprises worldwide. It fills the industry gap in large-scale, high-precision, full-scenario human motion data, empowering breakthroughs in core technologies such as full-body motion control for humanoid robots, dexterous hand manipulation, and human-robot collaborative interaction.

In addition, CHINGMU, together with Tsinghua University and Shanghai Film Art Academy, has jointly built China's first professional dance database, which is now officially listed on the Shanghai Data Exchange (SDE). Captured entirely using CHINGMU's high-precision optical motion capture system and its proprietary CMAvatar software, the database covers multiple core dance genres including classical Chinese dance and folk dance. To date, it has accumulated over 4,200 video clips with AI-annotated tags and a vast amount of high-quality human motion data, directly supporting applications such as digital human animation, AI model training, and content creation for film and games.

From high-precision optical motion capture systems, to multimodal Embodied AI data acquisition platforms, to openly shared data resources – CHINGMU is transforming its years of accumulated motion perception expertise into data infrastructure capabilities for the Embodied AI era. We believe that opening up data acquisition capabilities holds greater long‑term value than releasing a single dataset. CHINGMU looks forward to working with the OpenLET community and more industry‑academia partners to build a complete pipeline – from data acquisition and format adaptation to policy training – providing solid support for the continuous evolution of humanoid robot motion control capabilities.