Humanarchive provides the world’s largest egocentric multimodal dataset for embodied AI, capturing synchronized 3D pose, stereo vision, dense depth, tactile force feedback, and real‑time hand tracking from a wrist‑mounted point‑of‑view. The data, collected across diverse real‑world environments and annotated with action labels, object and scene segmentation, are accessible via a cloud platform with standardized APIs for easy integration into robotics, AR/VR, and autonomous agent training pipelines.
Funding
Funding not disclosed
Founders
Product
Problem
Developing embodied AI systems requires extensive, high-quality multimodal sensorimotor data that captures human actions from a first‑person perspective. Existing datasets are limited in scale, modality coverage, or realism, hindering progress in robotics, AR/VR, and autonomous agents.
Solution
Humanarchive creates the world’s largest egocentric multimodal dataset by capturing synchronized motion, vision, depth, tactile feedback, and hand tracking across diverse real‑world environments. Contributors use custom rigs equipped with 3D pose capture, stereo cameras, depth sensors, and tactile gloves to record natural human activities in settings such as homes, industrial sites, and service venues. The collected data are annotated with action labels, object and scene segmentation, and dense depth maps, providing a comprehensive sensorimotor record for training embodied intelligence. Researchers and developers can access the dataset through a cloud platform that supports standardized APIs for downloading raw streams and processed annotations, enabling rapid integration into machine‑learning pipelines.
Target Audience
Primary customers are robotics research labs, AR/VR developers, and AI companies building embodied agents that require large‑scale, multimodal human motion data for training and evaluation.
Features
- Synchronized capture of 3D pose, stereo vision, dense depth, and tactile force feedback from a wrist‑mounted point‑of‑view
- Real‑time hand tracking and high‑resolution tactile glove data for fine‑grained manipulation modeling
- Action labeling, object segmentation, and scene segmentation applied across all modalities
- Dataset spans over 50,000 contributors and 1,000+ custom rigs deployed in 125+ national partnerships across varied domains (e.g., construction, hospitality, healthcare)
- Cloud‑based access with standardized download APIs and metadata schemas for seamless integration into training pipelines
- Continuous expansion pipeline that adds new environments and task categories to keep the dataset up‑to‑date