In a factory workshop, a humanoid robot is sent to an automobile production line to sort parts, while at the other end the company's founder talks of having it move into ordinary homes. Moving robots from the exhibition stand onto the production line took Figure several years.
But the door from the production line into the living room has not really been pushed open. Machines already rolling off the line, models still on the drawing board, and the home written into the vision are three things at quite different stages, and mixing them up is the easiest way to read the story wrong.
Machines Rolling Off the Line
According to The Paper and Cailian Press, Figure's third-generation model is already handling logistics sorting at BMW's plant in Spartanburg in the United States, tidying mixed parts into sequencing carts that an automated transport system then takes to assembly stations. It is a real, repeatable job.
The signal from mass production is just as plain. According to company disclosures and several media reports, cumulative output of the third-generation robot has reached the thousand-unit mark, with the production beat rising from one robot a day early on to roughly one an hour. For a company only a few years old, that looks more like a production line that can actually turn.
What matters at the thousand-unit mark is not the number itself but the supply chain, production execution and quality consistency repeatedly proven behind it. Only when a robot is built enough like an industrial product rather than a laboratory exhibit will customers and capital treat it as a business worth serious study.
According to industry trackers, BMW's cooperation with the company began with an earlier model, first running long-cycle handling validation and then switching to the newer third generation for logistics sequencing. The carmaker still describes it externally as a validation phase and has not disclosed how many robots are on the job.
The Fourth Generation Still on the Drawing Board
On the hardware side, according to The Paper, the fourth-generation model has completed design lock, with components entering delivery and validation. Note the words design lock, not released; it is closer to a drawing that has been signed off, still some way from standing in front of a customer.
According to Cailian Press, the company positions the fourth generation as a standardized hardware base for a mix of general homes and complex industrial settings, wanting the model to carry out long sequences of tasks steadily in more unfamiliar environments. Converging the design means effort shifting from piling on new features to doing the settled things thoroughly.
The model side is equally restrained. According to The Paper and Tencent News, the company says its next-generation model is still in internal testing, with no benchmark details, paper or release date given; to the public, for now, it is only a preview, not a product that can be bought.
Feeding It Video From Around the World
The other card the company is playing is a data-collection project built around a smartphone app for ordinary people. According to The Paper and Cailian Press, it invites people to film everyday actions at home, in offices and in warehouses, then turns the videos into material for training robots.
As of that internal all-hands meeting, checked by several outlets, the project had gathered about 37.5 million clips of human activity, covering more than a hundred countries. Casting the net wide is about variety, meaning different people, different settings and different objects, rather than repeated motions inside a single factory.
The logic is not hard to follow: real-world tasks vary endlessly, and it is impossible to collect data and train a separate model for every job. Pre-training on a vast pool of human behavior and then adapting specific skills with small amounts of task data is what the company sees as the main road to making robots smarter.
The cost of that road is in the reporting too. According to Cailian Press, the company signed a computing partnership of substantial initial size for it, with data spending running into millions of dollars a week. Whether scale and variety are truly the right answer still has to be proven by how the models perform in unfamiliar settings.
According to The Paper, a new model pre-trained on this data completed whole tasks such as tidying, folding and making beds in previously unseen homes at a markedly higher rate than before. What all that variety is actually worth is starting to have results that can be compared.
Endurance on a Treadmill
There is one telling detail. According to overseas media and industry trackers, a third-generation robot once ran for nine straight days on a treadmill at company headquarters, as an endurance stress test for its legs and actuators; it was powered by a cable throughout, testing the mechanics rather than battery life.
It shows the hardware is sturdy enough, but it is far from proving a robot can work a shift on its own. Durability is one thing; working steadily through an unwatched shift and recovering from surprises is another, and between the two still lies the test of a real site.
According to overseas media, the company's founder said the roadmap had shifted greatly over the past year, with the focus moving to the fourth-generation model and volume production. That is a choice of direction, not a result already delivered, and it should be read apart from the machines rolling off the line in the workshop.
The Preorder Slip for the Home
According to the company's public product materials, the third-generation model was redesigned for home and commercial scenarios, adding close-range vision near the palms, wireless charging and a softer outer shell, with touch sensing in the fingertips, all to make it safer in spaces where people move about.
Figure is not the only one trying to bring robots into the home. According to Guangming Online and China Economic News, another company's home humanoid is priced at about US$20,000, or available by monthly subscription, with a soft outer shell and a two-legged design suited to doors, windows and appliances built for people.
A preorder is a preorder, and delivery is delivery. According to those same reports, the product is planned to begin deliveries in 2026, yet by the middle of that year there was still no independently verified record of an actual delivery, and the company also says early users will receive only basic autonomy.
For complex tasks it cannot manage, a remote expert can take over at an agreed time. That lets a robot finish a job sooner, but it also pushes privacy and labor costs to the fore: who can see inside the home, and when, are questions only real operations can answer.
Would It Still Recognize a Different Room
The hardest gate for the home is generalization. Succeeding in a prepared room is not the same as succeeding in a different room with a different cup; according to media reports, although the zero-shot home-task completion rate the company has published is improving, it is still a clear distance from steady reliability.
Before a robot can replace a person, it has to be validated again and again in defined settings. Factory inspection, warehouse handling and mall guidance have clear boundaries and room for error, making them the most realistic landing spots for now; as for the endless variables of millions of homes, the timetable is far from written.
Back to the door at the start. A robot rolling off the production line means it has cleared the first hurdle of being usable at all; but to become a roommate in your home, it still has to prove it recognizes a different room and makes no mistakes when nobody is watching. The dream is close, and it is also picky about the work.
Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.