Xperience-10M

Ropedia's open dataset of 10,000 hours of first-person, multimodal human experience, aimed at embodied AI.

Xperience-10M is an open dataset from the physical-AI data company Ropedia, published on Hugging Face according to the company's blog in March 2026. The data was recorded by people wearing a multi-camera capture device in real daily life; the “10M” in the name refers to 10 million “experiences” (one complete interaction), totaling 10,000 hours. Every experience synchronously records 6 video streams (4 fisheye, 2 stereo), audio, stereo depth, camera pose and SLAM trajectory, hand motion capture, full-body motion capture, and IMU data, along with multi-level language annotations for task, subtask, action, interaction, and object — about 2.88 billion RGB frames in total, roughly 1PB of data, which the company describes as the largest first-person dataset with structured 3D/4D annotation. It can be used for pretraining world models, robot learning, and first-person perception. It's restricted to research and non-commercial use, requiring an application, review, and signed agreement to download; a separate enterprise version exists for commercial use.

ExampleA researcher pretrains a hand-motion prediction model on the dataset's hand motion-capture and first-person video, then fine-tunes it on a small amount of robot data.

Also called

Ropedia Xperience-10M

Related

Egocentric Video · Human Video Data · Wearable Data Collection · Ego4D · World Model · Motion Capture

Sources

ropedia-ai/xperience-10m (Hugging Face)
Xperience-10M Dataset (Ropedia Blog)

As of

2026-03