TL;DR: We’re capturing the structure of how humans interact with the physical world to model sensorimotor intelligence at scale
Launch Video: https://www.youtube.com/watch?v=pFHr3GlTpck

The Problem:

Modern AI is largely a knowledge transfer problem. Over the past decade, training on trillions of tokens of internet data enabled breakthroughs in LLMs, diffusion models, and vision-language models. But human intelligence is predominantly embodied, and the internet does not capture it. Every day we manipulate objects, apply force, and subconsciously adapt in noisy real-world environments. Yet there is no dataset that captures the structure of human interaction with the physical world at scale. As a result, progress in **embodied spatial intelligence is bottlenecked by data.**

What we built:

Over the past two months, we built infrastructure for high-quality data collection at scale, including:

• Custom hardware rigs
• Internal models for policy benchmarking
• QA pipelines, alignment, and annotation software
• 25-person operations team
• Dedicated servers at our AWS data center for terabyte-scale offloading

Now that the system is live, we can collect up to 8,000 hours of data per day and we’ve signed national level partnerships to scale our contributor network to 50,000+ people. We’ve already shipped datasets to frontier research teams, and have built the largest multimodal dataset of it’s kind.

Our datasets:

We are currently collecting data across homes, restaurants, hotels, retail, transportation, construction, horticulture, and industrial environments across two datasets.

HA-Multi is a fully aligned multimodal dataset with vision, stereo depth (IR dot projection), tactile gloves, body IMUs, and wrist cameras. For customers, we provide structured outputs and visualizations including 3D MANO hand reconstructions, 2D tactile force maps, depth maps per timestamp, and human pose reconstructions.

HA-Ego is a mono RGB vision and wrist cameras dataset.

We provide annotations and metadata including environment and scene descriptions, high-level task descriptions, task-aligned atomic action labels, hand tracking, object segmentation, SLAM (extrinsic and intrinsics), and 3D pose reconstruction.

Team:

Hi, we’re Shloke, Samay, Rushil, and Raj. We’re engineers from Stanford and Berkeley who’ve spent our entire lives building across operations, hardware, and robotics. We’ve known each other for over 20 years and joined forces to dedicate the rest of our lives to building the Common Crawl for human sensorimotor intelligence.

Our Ask

We'd love to talk with you if:

• You’re building in robotics, world models, or evaluations
• You work at or with frontier AI labs interested real world
• You know cleaning companies, hotel chains, restaurants, factories, or other businesses interested in scaling efficiency and profits with AI

👉 Contact us at raj@humanarchive.ai

💼 Follow us on LinkedIn: https://www.linkedin.com/company/human-archive/

📱 Follow us on X: https://x.com/babugi28