Figure AI, an American robotics company, is once again raising the bar for humanoid robotics media attention with a demonstration claiming that a team of its humanoid robots has worked for a full 8-hour shift at human performance levels. According to a company announcement, the operation was conducted entirely autonomously using Helix-02, their AI system for full-body robotic control. The demonstration focused on an industrial/logistics task: identifying, manipulating, reading codes on, and placing packages onto a sorting conveyor belt. This is much closer to a warehouse or factory environment than a simple lab demo.
However, this test went significantly further. Through a live stream on X, the company showed how several Figure 03 robots managed to accumulate up to 24 consecutive hours of work, manipulating over 30,000 packages. The key is that Figure is not just presenting a robot performing a single action, but a coordinated fleet capable of maintaining operations throughout a standard workday. This is important because many humanoid demonstrations are fleeting, lasting only seconds or a few minutes, are highly constrained, or rely on carefully prepared conditions. Clearly, this is a public demonstration proving their robots are suitable for replacing humans in this type of work.
Helix-02 Aims to Replace Humans in Repetitive Work Tasks
Helix-02 is described by Figure AI as an evolution of its Helix system, capable of controlling the entire humanoid body. This encompasses sensory data, combining vision, touch, proprioception, locomotion, and balance, all within a unified architecture. The company had previously presented Helix-02 as a leap forward from systems where walking, balancing, and object manipulation are treated as separate problems. According to Figure, the system connects onboard sensors to actuators via a visuomotor network and introduces a full-body controller trained with over 1,000 hours of human motion data and reinforcement learning.
This aligns with the design of the Figure 03, the company’s third-generation humanoid. The robot was introduced as a platform specifically designed for Helix, featuring higher-speed, lower-latency cameras, an expanded field of view, cameras in the palms of the hands, and tactile sensors capable of detecting very small pressures. In theory, this setup allows the robot to maintain visual and tactile control even when its primary cameras are partially obstructed, such as when reaching for objects inside a container or manipulating parts in tight spaces.
The most compelling aspect of the 8-hour demonstration is not just autonomy, but operational endurance. Figure’s F.03 battery has a capacity of 2.3 kWh, and the company states about 5 hours of autonomy at maximum performance, with 2 kW fast charging and active cooling. This implies that an 8-hour shift likely relies on fleet management rather than a single robot working non-stop. It involves robots entering and exiting service, pausing for recharging, unit replacement, or task rotation. In fact, Figure had previously argued that the Figure 03 could operate almost continuously if it could return to a charging base during certain intervals.
Demonstration Included Humorous Moments Where the Figure 03 Robot Made Mistakes
This demonstration also arrives at a time when Figure is striving to prove that its humanoids are more than just promotional videos for social media. In 2025, the company had already shown results from its deployment with BMW. At the Spartanburg plant, Figure 02 robots worked 10-hour shifts from Monday to Friday, accumulating over 1,250 operating hours, moving more than 90,000 parts, and participating in the production of over 30,000 BMW X3s. BMW later confirmed that this deployment served as a precedent for further humanoid robotics testing in production.
The difference is that the BMW deployment was more tied to a very specific task of manipulating metal parts. Now, Figure’s current message points to something more ambitious: humanoids capable of working as a team, for a full shift, with a generalized AI that doesn’t depend on rigid programming for every movement. This is what the sector is aiming for: transitioning from the classic industrial robot, which is fast but fixed and specialized, to a more flexible humanoid that can be integrated into human-designed spaces without completely redesigning the factory or warehouse.
Despite these advancements, significant information is still missing. We know that in those 24 hours, Figure’s robots handled over 30,000 packages, inevitably with errors along the way. This included packages not being turned correctly, and even some ending up on the factory floor. However, these are minor issues, and it’s clear that only a couple of questions are truly important: First, how many units were needed to match a human worker? And second, what was the energy cost, to draw a conclusive proof of economic or productive parity? But it is clear that in the long run, these robots working as a coordinated fleet, 24 hours a day, 7 days a week, and 365 days a year, sounds very tempting from a numerical perspective.
