Physical AI moves from demonstrations into real manufacturing.
Beijing | July 2026
Humanoid robots are beginning to perform tasks once reserved for human workers inside automotive factories, marking a significant transition from controlled laboratory demonstrations to active industrial production.
Two humanoid robots recently completed a three-hour trial inside Xiaomi’s electric-vehicle factory, performing assembly and material-handling duties while attempting to keep pace with one of the world’s most highly automated automotive production lines.
During the test, the machines installed nuts, transported components and completed approximately 90 percent of their assigned work. The most demanding requirement was not simply executing each movement correctly, but maintaining the speed required by a factory capable of producing a finished vehicle every 76 seconds.
That figure does not mean one humanoid independently builds an entire automobile in just over a minute. Xiaomi’s production rate results from an integrated industrial system involving more than 700 conventional robots, automated transport platforms, artificial-intelligence controls and specialized machinery distributed across multiple manufacturing stages.
The humanoids entered this broader ecosystem as experimental workers rather than complete replacements for the existing production system. Their performance nevertheless demonstrated that human-shaped machines can operate alongside fixed industrial equipment without immediately slowing the line.
Traditional factory robots are usually designed for one highly specific activity. They weld, paint, lift or position components from a fixed location, often behind protective barriers. Their mechanical structure is optimized for a repetitive task rather than for mobility or adaptation.
Humanoid robots follow a different logic. Because factories, tools, aisles and workstations were originally designed around the human body, a machine with two arms, two legs and dexterous hands may be able to move between tasks without requiring the entire workplace to be rebuilt.
This versatility is one reason automotive manufacturers are investing heavily in physical AI, a field that combines artificial intelligence with machines capable of perceiving and acting within the physical world.
The robots use cameras, force sensors and vision-language-action systems to identify objects, interpret instructions and convert digital information into movement. Instead of following only a rigid sequence programmed in advance, advanced models can adjust their actions when a component changes position or an obstacle appears.
Their current abilities remain limited. A 90 percent completion rate may be impressive for an experimental deployment, but automotive manufacturing demands extremely high reliability. A repeated failure rate of 10 percent would be unacceptable across thousands of daily operations unless human workers continued supervising and correcting the machines.
Speed is another critical obstacle. A robot may successfully install a component during a demonstration but still lack the consistency required to perform the same task continuously over an eight- or ten-hour shift.
Battery life, heat management, maintenance and safety also determine whether a humanoid can become economically useful. Machines working near people must respond predictably to unexpected contact, dropped objects or sudden changes in the production environment.
Xiaomi is not alone in pursuing this transformation. BMW has already used Figure AI’s humanoid robots at its Spartanburg plant in South Carolina.
Figure 02 contributed to the production of more than 30,000 BMW X3 vehicles during an extended deployment. It completed approximately 1,250 hours of operation, moved more than 90,000 sheet-metal components and performed around 1.2 million steps.
Its principal task involved retrieving metal parts and positioning them with millimetre-level accuracy before welding. The operation was repetitive and physically demanding, making it an appropriate early application for humanoid automation.
BMW has since introduced Figure 03 into a more complex logistics workflow. The newer robot must manipulate components while repositioning its body, handling carts and responding to changes within an active manufacturing environment.
The progression reflects the industry’s strategy. Companies are beginning with predictable operations before attempting activities requiring greater judgment, dexterity and coordination.
Hyundai is preparing to introduce Boston Dynamics’ Atlas into its Georgia manufacturing facility by 2028. Tesla continues developing Optimus for internal industrial use, while Toyota and other manufacturers are testing humanoids for assembly, logistics and material transportation.
The expansion is creating serious labor concerns. Workers fear that machines initially presented as assistants for dangerous or exhausting duties may eventually replace large numbers of employees once their reliability and cost improve.
Manufacturers generally argue that humanoids will reduce injuries, address labor shortages and allow people to concentrate on technical supervision and higher-value activities. The long-term employment outcome will depend on whether companies retrain existing personnel or use automation primarily to reduce payroll.
Some jobs will likely disappear, particularly those based on predictable physical repetition. Other positions may emerge around robot maintenance, simulation, training, cybersecurity, fleet management and industrial safety.
The transition could also alter entry-level employment. Factory workers have traditionally gained experience through basic material handling or assembly duties before progressing toward technical positions. When robots absorb those initial tasks, companies may need new pathways for developing human expertise.
Economic calculations will ultimately determine the scale of adoption. Humanoids must produce enough value to justify their purchase, energy consumption, software subscriptions and maintenance.
A human worker can recognize unusual sounds, detect damaged materials and improvise when equipment fails. Replicating that situational awareness remains one of robotics’ most difficult challenges.
Highly automated plants also introduce systemic risks. When one human worker makes a mistake, the effect may remain limited to one workstation. When a centralized digital system fails, disruption can spread across an entire production line.
Cybersecurity becomes equally important because connected robots can represent entry points into industrial networks. A compromised system could interrupt production, damage equipment or expose sensitive manufacturing information.
Despite these limitations, the direction of development is clear. Humanoid robots are no longer confined to promotional videos showing them walking, dancing or carrying boxes under laboratory conditions.
They are entering factories where performance is measured through cycle time, precision, reliability and cost. Xiaomi’s three-hour experiment does not prove that human workers have become unnecessary, but it demonstrates that humanoids can begin participating in production moving at industrial speed.
The automobile rolling off the line every 76 seconds remains the product of an extensive automated system supported by human engineering and oversight. The emerging question is how much of that human involvement will remain necessary as physical AI becomes faster, more independent and more capable.
Phoenix24 | Automation advances when machines learn the rhythm of work. La automatización avanza cuando las máquinas aprenden el ritmo del trabajo.