Problem

Robotics and physical-AI companies need vast amounts of real-world demonstrations to build dexterous, perceptive systems that can generalize beyond controlled environments. Internet-scale text has advanced knowledge-based AI, but it does not provide the synchronized visual, tactile, depth, motion, and audio data needed to teach robots how humans perform physical tasks. The killer use case is training robotics foundation models on everyday work across homes, restaurants, hotels, retail, construction, agriculture, and industrial settings.

The pain is both scarcity and cost: collecting this data manually requires hardware, operations, workers, consent, annotation, and quality control at scale. Human Archive lowers the collection cost by using existing service and gig-work networks; workers are paid a base rate of about $1 per hour for participating, while consumers can receive discounted services in exchange for consenting to recorded visits. This creates a potentially scalable supply model, although privacy, consent, worker compensation, and dataset consistency remain important execution risks.

Product / Service

Human Archive is a multimodal data lab and infrastructure provider. It builds custom camera-and-sensor rigs, deploys them through workers performing ordinary tasks, and collects aligned egocentric video together with stereo depth, tactile, audio, body-IMU, motion-capture, and wrist-camera data. Its operations layer includes contributor networks, partner businesses, custom hardware, dedicated data infrastructure, and internal QA, anonymization, alignment, and annotation pipelines.

The company delivers structured, high-fidelity datasets and visualizations to frontier robotics labs and general-purpose robotics companies. Outputs can include 3D hand reconstructions, tactile force maps, timestamped depth maps, and human-pose reconstructions, giving customers training and evaluation data they would otherwise have to collect themselves. Human Archive also says it is developing internal model fine-tuning and robot-evaluation capabilities to demonstrate dataset quality and task effectiveness.

Market

Human Archive competes in the emerging physical-AI and robotics-training-data market, specifically multimodal embodied-intelligence data and world-model infrastructure. Its alternatives include Scale AI's robotics-focused data-labeling business, robotics companies' internal data-collection efforts such as Physical Intelligence's, Apptronik's data initiatives, and other specialized robotics-data startups. The broader demand is being driven by humanoid and general-purpose robotics companies racing to train foundation models.

The company is active rather than merely conceptual: it reports more than 1,000 active headsets, capacity of up to 8,000 collection hours per day, a contributor network being scaled to 50,000-plus people, national-level partnerships, and datasets already shipped to frontier research teams. Human Archive also announced an $8.2 million seed round from Wing Venture Capital, NVP Capital, Y Combinator, and prominent AI-industry angels. Public sources do not disclose revenue, pricing, or named enterprise customers, so it is best described as an early commercial-stage company with operational traction and undisclosed revenue status, not definitively as pre-revenue.