Figure AI Launches Index App Crowdsourcing Human Video to Train Robots, Commits $1 Billion to Data

Figure AI publicly launched Index on August 25, a consumer-facing app and data-collection platform previously developed under the stealth initiative Project Go-Big. The platform has amassed 16 million video uploads across 108 countries, processing 30 minutes of footage per second from more than 44,000 weekly active users. Figure has distributed $15 million in payouts to contributors and committed over $1 billion to data acquisition and compute over the next 12 months, founder and CEO Brett Adcock said in an announcement. Index is rolling out on iOS and Android alongside the public announcement.

Why Crowdsource Human Video

Index is Figure's answer to a fundamental robotics data problem. While language models scale on the public internet, robotics has long struggled with a severe lack of embodied interaction data. Standard methods — direct robot teleoperation, kinesthetic teaching, and synthetic simulation — remain slow, labor-intensive, and difficult to scale across varied environments. Index operates as a two-sided platform: individual "Creators" record themselves executing everyday household or workplace tasks for cash bounties, or users book gig workers through the app to complete chores on-site while capturing first-person footage.

The Ingestion Pipeline

Ingesting continuous consumer-generated video at high volume introduces major quality-control hurdles. Figure built an automated five-stage ingestion pipeline: filtering for resolution, frame rate, lighting, and task relevance; fraud review with dedicated human auditing teams; deduplication using high-dimensional vector embeddings to identify redundant footage; rebalancing using task quotas and embedding clusters to prevent over-representation of simple tasks; and hierarchical annotation that aligns high-level task descriptions with low-level physical manipulations for multimodal model training. Per 1,000 hours logged, Figure says the dataset captures 373 unique tasks, 1,146 distinct manipulated objects, and 116 unique environments.

Commercial Strategy Context

The Index launch follows Figure's recent milestones: the company's 1,000th Figure 03 build at its BotQ facility in San Jose, the deployment of Figure 03 units at BMW's Spartanburg plant for logistics sequencing, and an automated sorting pilot with Catalyst Brands in Reno. Adcock has argued that hardware scalability is no longer the primary hurdle — the true bottleneck is onboard intelligence and general-purpose reasoning. By scaling Index into a multi-million-dollar data-collection engine, Figure is attempting to build the equivalent of ImageNet for embodied AI.

What to Watch Through Year-End

Three checkpoints follow. Whether human video alone can provide the tactile feedback and force dynamics needed for intricate physical manipulation — the central research question Helix 02 will be evaluated against — will determine the long-term value of the Index dataset. The pace at which Helix 02's performance improves as Index data flows into training runs, expected to be visible in quarterly capability demos through Q4 2026, will validate the $1 billion data-and-compute bet. And Figure's competitors in the data-for-embodied-AI race — including Physical Intelligence, Skild AI, and Genesis AI — will publish their own data-collection approaches in the coming quarters, defining whether Index becomes the industry standard or one of several competing pipelines.