Figure Launches Index App to Train Humanoid Robots

Figure has officially launched Index, a mobile app previously run in stealth that pays contributors to record video of everyday physical tasks. The app has already amassed 264,000 downloads across 108 countries, drawing 44,000 weekly active contributors who have uploaded 16 million videos. This crowdsourced engine ingests 30 minutes of video every second, yielding roughly 4.9 years of human activity daily. To date, Figure has paid $15 million to these creators and plans to spend over $1 billion on data and compute over the next year to expand the dataset a hundredfold.

The resulting dataset provides the diverse physical training data required by Figure's Helix vision-language-action model. For every 1,000 hours of footage, Index captures 373 unique tasks, 1,146 unique manipulated objects, and 116 unique environments. To process this massive influx, Figure built a five-stage pipeline. It filters for quality, audits for fraud, uses embedding-based deduplication to reject repetitive clips, applies task quotas to rebalance the dataset, and generates hierarchical text captions.

This crowdsourced strategy bypasses the slow, expensive process of using teleoperation rigs or lab demonstrations. It suggests that egocentric human video, captured on standard smartphones, can serve as a viable pretraining signal for robot foundation models. For AI practitioners, Figure's pipeline offers a blueprint for data curation, proving that embedding-based deduplication can successfully incentivize contributors to submit novel tasks rather than spamming repetitive actions.

The initiative places Figure at the forefront of a highly capitalized race to train its Figure 03 humanoid against rivals like Skild AI, Physical Intelligence, and Atoms, the latter of which recently closed a $1.7 billion funding round. Alongside its Brookfield partnership to collect video across millions of square feet of real estate, Figure aims to leverage Index as a future marketplace. The company envisions transitioning the app from paying humans for chores to dispatching humanoid robots to perform those same services.