10,000-Hour Egocentric Full-Body Multimodal Dataset Purpose-built for fine-grained embodied AI manipulation training, spanning diverse real-world scenarios with high-definition stereoscopic video, full-body joint poses, and high-density semantic annotations.

For more details, please refer to the link: https://www.nexdata.ai/datasets/embodied-ai/2236?source=Opendatabay

10,000-Hour Egocentric Full-Body Multimodal Dataset

Covers residential, retail & office scenarios (kitchen, bedroom, living room, supermarket, office) with diverse real-life tasks: meal prep, cleaning, storage, garment care, merchandising & picking

Each sample includes spatiotemporally aligned 4K stereo video, camera calibration params, 76-point full-body pose & step-by-step annotations

Adopts PICO 4 Ultra head-mounted stereo camera + wrist & ankle IMU motion capture solution

Supports dense semantic & action-level annotations; all data passes multi-stage quality control reviews

Supports 4096×1536 / 30fps HD video output, tracks 24 torso joints and 52 hand joints, with frame-wise dense annotations and full-process quality control

Commercial License