發生了什麼事
Glitchwire reports that Figure AI announced Index, a crowdsourced platform designed to collect videos of people performing everyday physical tasks for humanoid-robot training. Figure says the platform accumulated 16 million uploads during a four-month stealth period and is paying contributors for data. The report says Figure plans to spend more than $1 billion on data and over the next 12 months.
Glitchwire reports that Figure AI has brought Index out of stealth after four months of operation. According to the report, Figure says the platform has received 16 million video uploads from users in 108 countries, passed 264,000 app downloads and reached more than 43,000 weekly active users. Figure also says it has paid contributors $15 million and that uploads are arriving at a rate of more than 30 minutes of video per second. These figures come from Figure’s announcement and company-linked material; the supplied report does not independently verify them.
The platform’s purpose is to gather demonstrations of physical tasks for humanoid robots. Glitchwire describes examples including opening drawers, folding laundry and navigating cluttered spaces. The report says Figure views this information as a missing counterpart to the internet text used to train language models: physical-world demonstrations are harder to collect at scale and include environmental variation that simulations or controlled teleoperation may not capture. Index’s approach is to pay ordinary people to record everyday activities in their own environments.
Glitchwire places Index within Figure’s broader effort to train its Helix system, which the report says uses human demonstrations captured through teleoperation, robot-performance logs and large-scale video of people doing everyday tasks. The article also describes Figure’s Project Go-Big initiative and a partnership with Brookfield involving residential units, but it presents Index as a separate crowdsourcing strategy. Figure says it intends to spend more than $1 billion on data and during the next 12 months. The report says the platform is live and accepting contributors, but it does not provide independently verified terms of participation or a public technical evaluation.
為什麼這很重要
Index targets a central limitation of embodied AI: robots need large amounts of real-world task data, including examples from varied homes, environments and materials. If Figure’s reported scale and data quality hold, the platform could influence how humanoid robots are trained and how companies recruit people to provide training data. Its longer-term labor implications remain uncertain.
The significance of Index is that it treats data collection as an infrastructure problem for physical AI rather than only a robotics-hardware problem. A robot operating outside a factory must deal with different layouts, surfaces, objects and household routines. Glitchwire reports that collecting contributions across 108 countries is intended to expose Figure’s systems to cultural, architectural and material variation. That could help address the long tail of situations that are difficult to reproduce in simulation, although geographic breadth alone does not demonstrate that the data is representative or useful.
The project could also affect the economics of robot development. Figure’s reported $15 million in contributor payments indicates that the company is already paying for data at meaningful scale. If the videos are sufficiently labeled, diverse and relevant to the actions robots must perform, crowdsourcing could become a repeatable way to expand training sets. But the report does not explain how contributors are selected, what tasks they are asked to perform, how quality is checked, or how privacy and consent are handled when recording inside homes. Those omissions limit what can be concluded about the dataset’s value.
The labor implications are potentially substantial but remain prospective. Glitchwire says Figure explicitly connects Index to a future service in which customers lease robots for household work, and it reports that CEO Brett Adcock has discussed a possible monthly lease price of roughly $400 to $600. The article also says Figure’s robots supported production of more than 30,000 BMW X3 vehicles over 11 months in South Carolina. Those claims are not independently confirmed in the supplied material, and the report offers no evidence that a household robot can currently perform a broad range of tasks reliably. The immediate development is a data-collection product, not proof that household labor has been replaced.
互動機制：它實際上是如何運作的
以互動方式探索這項發展背後的基礎技術。
An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
接下來看什麼
The key questions are whether crowdsourced videos translate into reliable robot behavior, how contributors are paid and protected, and whether Figure can scale the model economically. The report does not independently confirm Figure’s usage, payout or performance figures, and it provides no public evaluation showing that Index has improved a deployed robot’s capabilities.
The first test is technical: whether videos of humans performing tasks can improve robot control in real environments. Useful evidence would include independently reproducible evaluations, clearly defined tasks, failure rates, across homes and objects, and comparisons with teleoperation or simulated training. Glitchwire reports that Figure has demonstrated robots in controlled industrial settings, but it does not provide a new Index-based result or a showing that the platform has changed performance.
The second test is economic. Figure’s current reported activity may not predict the cost of acquiring and reviewing data at ten or one hundred times the present scale. The report says the company is framing its path as a 100-fold expansion and that the existing $15 million in payouts suggests the model works at current scale. That is an assertion about present economics, not an independently verified forecast. Future reporting should examine contributor compensation, task availability, platform retention, data-cleaning costs and whether the resulting training data produces measurable commercial value.
The third test concerns governance and the distribution of benefits. A system that pays people to teach robots could create new income opportunities, but it could also turn ordinary household activity into a source of commercially valuable behavioral data. The supplied report does not state what rights contributors retain, whether recordings include bystanders or private information, how data can be withdrawn, or how Figure handles safety risks. It also does not establish whether any robot-service rollout will occur, at what price, or at what level of reliability. Until those questions are answered, Index is best understood as a significant data-collection launch with uncertain downstream effects.