Figure Robots Work at BMW as China Takes Most Robot Sales
Figure’s humanoid robots are working at BMW, while Tesla’s Optimus has been demonstrated performing household chores. Those two settings make the technology’s direction unusually clear: developers want one adaptable machine architecture to move from structured production lines into homes. They do not yet establish that humanoids can operate reliably at mass scale without close supervision.
The manufacturing race adds pressure to close that gap. China accounted for more than half of robots sold globally in 2023, reflecting an industrial base positioned to reduce hardware costs and expand production. That figure covers robots broadly, however, rather than humanoids alone. It is evidence of manufacturing scale and robot adoption, not a count of human-shaped machines ready for factories or living rooms.
A factory job is a narrower test
Figure’s work at BMW is significant because factories can provide conditions that suit today’s control systems. Components can arrive at defined stations, walking routes can be managed, and a task can be broken into repeatable motions. Engineers can also modify tooling, containers and workspaces around the robot instead of expecting the machine to handle every possible variation.
That makes industrial deployment more demanding than a staged demonstration but more constrained than general household work. A production robot must repeat its task for useful periods, fit the line’s cycle, stop safely around people and recover when an object is displaced or a grip fails. A successful manipulation sequence shows sensing and actuation working together; dependable deployment requires the entire system to handle exceptions without repeatedly calling for human help.
Humanoids typically combine cameras and microphones with multimodal models that process objects, images, text, video and spoken language. Those inputs can help a robot interpret a request and select an action. Locomotion controllers must then maintain balance, while arms and hands turn that decision into physical movement. Errors can accumulate across that chain: recognizing a garment is different from finding a stable grip, separating tangled fabric and recovering when it slips.
This is why Tesla’s household demonstrations and its industrial ambitions represent different readiness levels. Optimus has been shown folding a shirt, but completing a load of laundry would require it to identify mixed garments, separate them and adapt its grip and folding sequence. In another public demonstration, a human remotely controlled Optimus while it handed out snacks. Teleoperation can demonstrate the mechanical platform’s reach and grasping, but it does not show that the robot can independently plan and recover from the same task.
China’s scale changes cost before it proves autonomy
China’s broader robotics position could still materially affect humanoids. According to International Federation of Robotics data presented by Our World in Data, China installed 276,000 industrial robots in 2023, twelve times its 2011 total. Its installed base reached about 1.76 million robots that year. By comparison, annual US installations rose from 21,000 in 2011 to 38,000 in 2023.
Those numbers describe automatically controlled, reprogrammable industrial machines, not specifically humanoids. Their relevance lies in the surrounding production system: suppliers, motors, drives, sensors, power electronics, batteries, machining capacity and integration expertise. Higher volumes can reduce component costs and improve manufacturing consistency, advantages that Chinese humanoid producers may be able to use even though conventional industrial robots remain a different product category.
Scale cannot manufacture its way around every deployment constraint. A humanoid sharing space with workers needs validated protective functions, controlled operating zones and defined responses to blocked routes, lost communications and maintenance interventions. Navigation sensors are not automatically safety-rated sensors, and changing a load or workspace can alter the hazards even if the robot’s software remains unchanged.
The comparison therefore favors factories as the first proving ground. Figure’s BMW work puts humanoids into an environment where performance can be measured against a specific task, while Optimus household demonstrations point toward a much less predictable destination. China’s robot-production scale could make the hardware cheaper more quickly, but mass household adoption still depends on something production totals cannot demonstrate: consistent autonomy when objects, people and surroundings refuse to stay in their expected places.
| More aerospace and engineering stories, right in your MSN feed. Follow AMI on MSN |
By Jonathan Barrett — Editor for AMI’s future mobility and autonomous systems section, with two decades covering robotics, e-mobility, drone-vehicle convergence, and transport mechanical systems.
