Field note
Figure Index makes diverse physical-world data collection a dedicated infrastructure layer in the humanoid-agent stack.
The launch matters because general-purpose robots cannot rely on internet text alone; they need broad recordings of human interaction with varied objects and environments. Figure says Index crossed 264,000 app downloads in more than 100 countries and 44,000 weekly active users during a four-month stealth period. Contributors uploaded more than 16 million videos, and Figure says the pipeline was processing thirty minutes of video every second at launch.
A consumer contribution app recruits and pays creators, applies quality controls, and turns diverse first-person physical activity into a proprietary dataset for training and improving Helix. Robotics teams should treat collection incentives, consent, labeling, distribution coverage, quality filters, and model feedback loops as core infrastructure decisions.
The scale figures come from Figure, and the company has not yet published enough independent evidence to connect dataset growth with general humanoid reliability. Revise the thesis when Figure releases dataset composition, consent and quality methods, controlled generalization results, and evidence that additional data improves deployed robot behavior.