Figure launches Index, a paid human-video pipeline for training humanoid robots

Why it matters

Humanoid developers face a shortage of varied real-world training data. Index gives Figure a global collection network, a potential proprietary moat and a way to test demand for robot-delivered services before its hardware reaches comparable scale.

Brett Adcock (@adcock_brett) launched Index on August 25th, turning Figure's four-month-old, crowdsourced data operation into a public app that pays people to record household and workplace tasks for its humanoid robots.

https://x.com/adcock_brett/status/2092303633559982106

In a thread on X, Adcock said Index had collected more than 16 million video uploads from 108 countries, paid contributors $15 million and reached 264,000 downloads while operating in stealth. Figure says the network now has more than 44,000 weekly active users and processes 30 minutes of uploaded video each second.

The scale claims are Figure's. Its Index announcement provides upload counts and throughput, though it does not translate the collection into a standardized total of accepted training hours. Figure says each 1,000 hours of retained data contains 373 distinct tasks, 1,146 manipulated objects and 116 environments.

Adcock founded Figure in 2022 after building recruiting marketplace Vettery and co-founding electric-aircraft developer Archer Aviation. In Figure's founding master plan, he framed humanoids as a decades-long, capital-intensive attempt to automate physical labor. Index extends that thesis into an unusually direct data operation: Figure is paying people around the world to demonstrate the work it wants its machines to perform.

A labor marketplace built to train Helix

Index has two sides. Contributors can apply to receive a recording device, capture first-person footage while completing approved tasks and earn money by the minute. Homes and businesses can also book creators to perform chores or work such as making beds, folding laundry, stocking shelves and serving customers.

Figure says the footage will train Helix, its vision-language-action system for controlling Figure humanoids. The company argues that internet video lacks the camera perspective, task detail and environmental coverage required to train a general-purpose robot. Index is designed to capture that long tail directly, including unfamiliar homes, objects and individual ways of completing the same job.

The iPhone app listing says demand to become a creator is high and applicants may be placed on a waiting list. Accepted contributors receive recording hardware and can browse eligible tasks and projected earnings. The listing shows the public Figure INDEX app reached version 1.0 three days before the announcement, consistent with Figure's description of August 25th as a rebranding and public launch after four months of stealth collection.

Index runs each submission through automated quality filters, fraud review, deduplication, dataset rebalancing and annotation. Figure says videos judged too similar to previously accepted footage are discarded, while the remaining clips receive hierarchical text captions for training.

That pipeline is the product underneath the app. Downloads and raw uploads are distribution metrics; Figure still has to convert the footage into data that improves robot performance across new tasks and environments.

Figure puts a $1B budget behind the data race

Figure committed to spend more than $1 billion on data and compute during the next 12 months, with Adcock saying Index is on a path to grow 100-fold. The combined figure does not separate contributor payments from model-training infrastructure, leaving the eventual economics of the creator network dependent on how Figure divides that budget.

The commitment is roughly the size of Figure's entire Series C financing, announced in September 2025 at a $39 billion post-money valuation. Parkway Venture Capital led that round, with Brookfield Asset Management, Nvidia, Macquarie Capital, Intel Capital, Align Ventures, Tamarack Global, LG Technology Ventures, Salesforce, T-Mobile Ventures and Qualcomm Ventures participating.

Figure had already previewed its data strategy through Project Go-Big in September 2025, when it described using human video collected in Brookfield properties to train Helix. Index expands the collection pool from managed properties and business partners to a consumer-style global network.

The approach addresses a basic constraint in humanoid robotics. Figure can manufacture additional robots, but robot-generated demonstrations remain expensive and limited by the number of machines available. Human contributors can produce video before Figure has deployed a comparable robot fleet. Figure has reported that its Helix 02 system used more than 1,000 hours of human motion data to train its whole-body controller, giving the company a direct use for a much larger corpus.

Index also carries a broader commercial option. Its privacy policy says collected information may include video, audio, location and characteristics derived from recordings. The policy permits Figure to disclose or sell personal data to commercial purchasers for purposes including research and product development, while allowing contributors to request an opt-out from such sales. That language leaves Figure room to commercialize parts of the collection operation beyond training its own robots.

Adcock's stated endpoint is a robot-as-a-service business. Index begins with humans arriving to clean homes or work inside restaurants, factories and logistics centers. Those workers generate the demonstrations Figure needs to train Helix, while the service network gives Figure a map of tasks customers are already willing to order. The human labor marketplace is therefore both the data engine and an early test of demand for the robots intended to replace it.