Figure AI Index Dataset: 16M Videos Scale Helix Training

Figure AI Emerges with Index to Solve Humanoid Data Bottleneck

Figure AI publicly launched its Index platform on August 25, 2026, revealing a consumer app that collected 16 million real-world human video uploads during four months in stealth. The dataset draws from 264,000 downloads and 44,000 weekly active users across 108 countries, delivering 30 minutes of new video every second—equivalent to 4.9 years of human activity uploaded daily. This approach directly addresses the shortage of diverse physical interaction data needed for general-purpose humanoid robots like the Figure 03.

The company shifted from purchasing third-party data after vendors failed to meet throughput, diversity, or quality standards required by its Helix AI models. Instead, Figure built an exclusive pipeline that pays creators by the minute to record everyday tasks in homes, restaurants, retail spaces, and logistics facilities. Per 1,000 hours collected, the dataset captures an average of 373 unique tasks, 1,146 distinct manipulated objects, and 116 unique environments, injecting the long-tail variation critical for real-world deployment.

This crowdsourced strategy supports commercial pilots, including logistics sequencing at BMW Group Plant Spartanburg, where Figure 03 units already operate. The data volume accelerates generalization in Helix, reducing the trial-and-error cycles that inflate development timelines and service overhead for early fleets.

Data Pipeline Details and Processing Throughput

Index videos pass through a five-stage pipeline of filtering, fraud detection, deduplication, rebalancing, and hierarchical annotation before feeding Helix training runs. The system handles extreme ingestion rates while maintaining quality, with 44,000 weekly contributors generating content that spans domestic chores like folding laundry and commercial workflows in stockrooms.

Figure reports that this volume already validates internal generalization improvements, with plans to scale 100x. The $15 million paid to creators to date represents only the initial outlay in a $1 billion commitment over the next 12 months for data acquisition and compute resources.

Such throughput directly lowers barriers to scaling robot fleets by shortening the time from data collection to deployable skills. Operators gain access to models trained on authentic variation rather than synthetic or narrow lab data, which historically leads to higher failure rates in unstructured environments.

Link to Figure 03 Production and BMW Deployments

The Index launch coincides with Figure scaling Figure 03 production past the 1,000-unit mark at its BotQ facility. These units incorporate Helix models trained on the growing dataset, enabling more robust performance in logistics and manufacturing settings such as the Spartanburg pilot.

Better training data reduces common failure modes like grasping inconsistencies or navigation errors in novel settings, which in turn extends mean time between interventions. This improves overall fleet availability and lowers the service burden on providers supporting Figure platforms.

The dataset's diversity across 108 countries also prepares robots for global variability in objects, lighting, and workflows, a factor that directly affects warranty claims and maintenance scheduling for international deployments.

Implications for Total Cost of Ownership and Robot-as-a-Service Models

Index positions Figure to offer robots as a service by accelerating the path to reliable autonomy. Lower training costs through crowdsourcing translate into more predictable operational expenses for end users, as robots require fewer on-site adjustments after initial deployment.

Service providers working with Figure 03 units will benefit from models exposed to millions of real demonstrations, potentially extending service intervals and reducing spare-part consumption tied to wear from repeated failures. The $1 billion investment signals sustained focus on data quality over hardware iteration alone.

Industry observers note that this volume of physical data creates a competitive moat, allowing Figure to iterate Helix faster than peers reliant on limited teleoperation or internet-scraped footage. For fleet operators, the result is faster ROI through higher uptime and reduced unplanned downtime.

Future Scaling and Contributor Ecosystem

The Index app now appears on Google Play and the App Store, enabling continued global growth beyond the initial 264,000 downloads. Figure plans to expand creator tools and business-side task booking, turning the platform into a dual data-and-labor marketplace.

Sustained collection at current rates would generate hundreds of millions of videos within months, further refining Helix for complex sequences in homes and factories. This scale directly supports the company's vision of robots handling the long-tail tasks that currently drive high service costs in early humanoid deployments.

By embedding data collection into everyday user activity, Figure lowers the marginal cost of each additional training example compared to traditional methods, a shift with clear downstream effects on maintenance economics and warranty structures for Figure 03 fleets.