Humanoid robots working live 24/7 are no longer just a lab experiment or a polished promotional video. The American company Figure AI has decided to put its Figure 03 units in front of a camera and let anyone see, in real time, how well they handle a full warehouse shift, relentlessly handling packages.
In this test, which is being streamed live, the humanoid robots work for hours on a parcel conveyor belt : picking up boxes, orienting them, releasing them, and repeating the process thousands of times. What's striking is not just the technology behind it, but the staging: an open broadcast , complete with timers and counters for processed parcels, transforming what was once a private tech demo into a small public event watched by hundreds of thousands of people.
A 24-hour workday at the level of human performance
Figure AI's initial idea was to demonstrate a full eight-hour shift under realistic warehouse conditions, using a Figure 03 humanoid controlled by their Helix-02 system. The official objective: to achieve what the company calls "human performance levels" in a repetitive package handling task, without breaks or direct operator intervention.
However, the demonstration fell short. After the first eight hours of operation, Brett Adcock, the company's founder and CEO, announced live that the experiment would continue for a total of 24 uninterrupted hours . The accumulated operating time and the total number of packages checked or handled by the various robots are superimposed on the broadcast.
In that first eight-hour period, the figures were already striking: more than 10.000 packets had been processed . Approximately 14 hours after launch, the counter was approaching 18.000. At other points during the broadcast, the counter reached over 34.000 packets in just over 27 hours of accumulated operation, in a run that the company describes as completely autonomous with Helix-02.
The live stream, available for free, has already garnered hundreds of thousands of views and thousands of interactions on social media. It's a simple scene: three humanoids—identified by names like Gary, Frank, and Bob on cards on their chests—in front of a treadmill, moving boxes without saying a word. But the constant rhythm and monotonous movements have a hypnotic quality that draws you in.
Figure AI itself presents it as a way to demonstrate resistance, work rate and sustained autonomy , precisely the ingredients that separate videos of dancing robots from a technology designed for real factories, logistics centers or warehouses where physical work lasts for hours.
Helix-02: the brain that moves the humanoid
Behind these images of robots handling packages lies a less flashy but crucial protagonist: Helix-02 , the full-body autonomy system developed by Figure AI to control its humanoids. More than just "off-the-shelf software," it's a neural network that coordinates the robot's vision, touch, proprioception, and actuators as if they were a single organism.
The company describes Helix-02 as a unified visuomotor network , capable of receiving direct information from all cameras, touch and position sensors, and translating it into coordinated movements of the arms, hands, legs, and torso. In other words, the robot's actions are not programmed step by step; instead, the system learns to move smoothly in different environments.
To train this comprehensive body controller, Figure claims to have used over 1.000 hours of real human movements , collected as baseline data, in addition to extensive digital simulations. In these simulations, the system tests actions repeatedly, corrects errors, and improves its performance before being deployed in the physical world.
According to the company, Helix-02 has allowed them to replace more than 100.000 lines of hand-written code, representing a shift in approach: instead of manually designing each movement routine, the neural network learns to control the robot's body with much greater flexibility. This is what makes possible scenes like watching a single humanoid completely unload and reload a dishwasher in a kitchen, integrating walking, crouching, manipulating dishes, and maintaining balance in a continuous task lasting several minutes.
In the specific case of handling packages in the live test, Helix-02 combines computer vision , integrated tactile detection and proprioception to locate each package, adjust the grip, correct the posture and repeat the cycle with a rhythm that, at many times, approaches that of a trained human operator, but without showing signs of fatigue.
Figure 03 and Helix 2: from the kitchen to the storeroom
The robots featured in these broadcasts belong to the Figure 03 family , electric humanoids approximately 173 centimeters tall, with an estimated payload capacity of around 20 kilograms and an approximate weight of 61 kilograms. Figure has showcased versions geared towards domestic tasks as well as variants focused on industrial environments.
At home, the company has shown how one of these robots, controlled by the Helix 2 system (the name by which its technology has also become popular), is able to load and unload a dishwasher autonomously , select small objects on a countertop, or manipulate utensils with considerable precision, something that until recently was especially difficult for humanoids.
On the industrial side, the 24-hour live test takes this same platform to a parcel sorting and picking environment . The video shows units like Gary and Frank operating on conveyor belts, picking up boxes of different sizes, orienting them correctly, separating them, and returning to the starting point without stopping.
One of the keys to this new generation is the use of advanced touch sensors and cameras in the palms , which allow the humanoid to perform actions that were previously virtually impossible, such as separating individual pills, dispensing with syringes, or distinguishing small and irregularly shaped objects from clutter, even when its own hands partially obscure the view of the main camera.
Although many of these capabilities are still considered preliminary results, the company maintains that the suite of tests is beginning to demonstrate what continuous body autonomy can offer in real-world tasks: sequences of dozens of linked actions — walking, turning, stretching, grasping, depositing — that are executed smoothly without constant human restarts or corrections.
Mass production: from one robot a day to one per hour
The live broadcast is not just a marketing ploy. It coincides with a key moment for the company: the leap from a pilot plant to a high-volume production facility . Figure AI has announced that it has gone from manufacturing approximately one robot per day to achieving a rate of one robot per hour in less than 120 days, representing a 24-fold increase in capacity.
At this new facility, the company claims to have over 150 connected workstations , more than 50 inspection points distributed along the assembly line, and around 80 functional tests at the end of the process before each humanoid robot is approved. It also states that it has already produced more than 9.000 actuators, the components that act as the robot's "muscles."
Another key statistic highlighted by Figure is its first-time success rate at the end of the production line, which, according to the company, already exceeds 80%. This is a promising percentage, although it also demonstrates that assembling humanoids remains a complex operation, with room for error and the need for adjustments even in such an automated environment.
In one of its latest updates, the company explained that in a single week they produced 55 units of their robots, a significant figure for a type of machine that until recently was manufactured almost entirely by hand. For the company, the combination of increasing mass production and continuous public demonstrations is a way to convince potential customers that these are no longer isolated prototypes, but a product destined for fleet deployment.
This approach has a clear interpretation for Europe and for Spain: if the production of humanoids enters a consolidated industrial phase, it is only a matter of time before European logistics companies, manufacturers and warehouse operators consider pilot tests or limited deployments in their facilities, at least for very specific tasks where the return on investment can be measured relatively quickly.
A live performance without filters: mistakes, limitations, and lessons learned
One of the most interesting aspects of the broadcast is that not everything goes perfectly . At various points in the video, you can see the humanoid fail to grasp a package, drop a box, or fall to the ground, and simply move on to the next item on the conveyor belt without picking up what it has lost.
Far from trying to hide these errors, the company leaves them visible on camera. This makes the demonstration more credible and, at the same time, more revealing: despite the advances, humanoid robotics continues to stumble over basic problems when faced with variations of the physical world that don't quite match what was planned in its training.
In logistics, the real challenge isn't simply picking up a standard, neatly placed box. The real headache comes when the package is twisted, dented, poorly stacked, slippery, or slightly out of position . It's in these situations that the humanoid still shows hesitation, somewhat erratic movements, or an inability to correct a failed maneuver.
The show, with its successes and shortcomings, also helps illustrate the difference between a machine and a person. A human operator is usually able to react to an exception : if a box falls, they pick it up, reorganize the workspace, and improvise a solution. The robot, for now, tends to continue with the repetitive task, ignoring the disorder it creates around it unless explicitly addressed.
Even so, the ability to work for hours on end without rest , maintaining a very constant pace and handling volumes of packages that would far exceed what is reasonable for one person in a single shift, makes it clear why this type of technology is being followed with such interest by sectors such as e-commerce, courier services or large-scale distribution, also in the European sphere.
Does it make sense to use a humanoid to move boxes?
The Figure AI demonstration has reignited a classic debate in the industry: does a robot need to have a humanoid form to work in a warehouse? Many industrial robotics specialists argue that if the goal is simply to handle packages, a machine designed specifically for that task is usually simpler, more stable, and probably cheaper.
For years, robotic arms and mobile platforms have existed for picking, packing, and palletizing tasks , operating without legs, heads, or complex human-like joints. Companies like Boston Dynamics market warehouse robots capable of handling packages weighing over 20 kilos, and logistics giants like Amazon have long been using their own robotic solutions in centers around the world.
So, where would the value of a humanoid like Figure 03 lie in a European logistics or manufacturing environment? The company's bet is on the flexibility of a human-like body : being able to walk, climb stairs, pass through standard doorways, use tools designed for people, and move around existing facilities without having to completely redesign warehouses, shelving, or workstations.
In other words, Figure's proposal is not limited to competing in the specific task of moving boxes, where other machines are already established, but rather to proposing that the same robot can assume different roles in varied environments: from warehouses and production lines to, in the medium term, maintenance services, assistance in physical stores or even advanced domestic tasks.
This approach has direct implications for countries like Spain, with a high concentration of small and medium-sized enterprises operating in facilities and workshops adapted to a human scale. In such cases, a humanoid robot capable of integrating into the existing space, without requiring a complete reconfiguration of the infrastructure, could prove more attractive than entirely redesigning the assembly line to accommodate fixed robots or rigid automated systems.
Work impact and social perception: from fear to curiosity
For years, the idea has been repeated that artificial intelligence and robots would eventually take jobs away from millions of people, especially those in office positions or those performing routine tasks in front of a computer. Faced with this scenario, many have suggested—sometimes half-jokingly, sometimes half-seriously—that the solution would be to become manual laborers.
The sight of several humanoid robots working a full 24-hour shift sorting packages, with productivity levels that any human would find extremely difficult to match, is a clear sign that repetitive physical jobs could also be affected in the coming years. The idea that manual labor was a safe haven from automation is beginning to crumble.
In the Figure broadcast, the counter shows how, in approximately 13 hours of continuous work , a single robot is capable of processing more than 17.000 packages. These figures force us to reconsider what kinds of tasks make sense for people to continue performing and which ones would be more logical to delegate to machines, both in warehouses and in other sectors.
The company itself uses everyday examples to illustrate this near future: from the robot that loads a dishwasher without help to the possibility that, one day, we will call an emergency "plumber" or "locksmith" and whoever appears at the door will be a humanoid capable of diagnosing the fault and repairing the installation with the precision of a veteran professional.
In Europe, where the debate on regulating artificial intelligence and protecting labor rights is particularly active, these kinds of demonstrations fuel the discussion. Governments, unions, and companies are beginning to consider scenarios in which humanoid robots share workspaces with people, not only in highly automated factories but also in urban service and logistics environments.
Meanwhile, the public's reaction to the broadcast of Figure is a mixture of fascination and unease . Many viewers acknowledge that the video is captivating and that it's impressive to see the robot working tirelessly, but at the same time, they wonder what will happen to the jobs that currently depend on the very repetitive tasks that humanoids are beginning to master.
Figure AI's move, by showcasing its humanoid robots working live 24 hours a day on warehouse tasks without any retouching, marks a turning point between spectacular demonstrations and the reality of industrial automation: with systems like Helix-02 controlling humanoid bodies already mass-produced, the question is no longer whether this technology will reach our factories, warehouses, or even homes in Spain and Europe, but rather how, at what pace, and under what rules it will be integrated into daily life and the labor market.