The Floor of Hall 52
In Hall 52 of BMW’s ten-million-square-foot assembly plant in Spartanburg, South Carolina, industrial automation is shedding its cages. For decades, the rule of automotive manufacturing was absolute rigidity: massive, multi-ton robotic arms bolted to the floor, executing highly precise, pre-programmed paths. If a stamped steel bracket arrived shifted 3 inches to the left, the sensors flared, the line halted, and a human operator had to intervene.
The arrival of the Figure 03, the latest general-purpose humanoid from California-based Figure AI, marks a departure from that deterministic engineering. Standing 1.73 meters tall and weighing 61 kg, the metallic frame operates directly alongside human workers. Rather than executing hard-coded loops, the machine adapts in real time to a shifting environment, fluidly moving between the delicate sorting of thin-walled components and the brute-force task of pulling heavy metal supply trolleys down the concrete floor.
From Lab Experiment to Factory Floor
This deployment represents the second phase of a deep commercial partnership between the German automaker and the Silicon Valley robotics firm. Throughout 2025, Figure’s second-generation machine, Figure 02, underwent a rigorous 11-month trial in the Spartanburg body shop. Tasked with the physically demanding job of loading sheet-metal parts into welding fixtures, that predecessor logged over 1,250 hours of operational runtime, handled more than 90,000 parts, and directly contributed to the production of 30,000 BMW X3 vehicles.
The trial proved that humanoid hardware could survive the grueling, multi-shift rhythms of a real factory floor.
Industry analysts and production experts analyzing the hardware shift note that the iteration cycle has accelerated. According to a comprehensive Figure 03 capabilities review, the mechanical and architectural refinements implemented in this latest version directly address the real-world friction points discovered during those thousands of hours on the factory floor.
“Our 11-month deployment of Figure 02 proved that humanoids are no longer lab experiments; they can be a valuable asset in establishing a flexible, reliable manufacturing workforce,” said Brett Adcock, founder and CEO of Figure AI.
With those operational miles in the bank, Figure has retired the second-generation fleet to its headquarters, channeling the hardware and software lessons directly into Figure 03.
Image Credits: Figure AISolving the Sequencing Trap
While the body shop trial focused on classic, localized pick-and-place manipulation, Figure 03 has been sent to the logistics hall to tackle an entirely different operational bottleneck: parts sequencing.
In automotive manufacturing, sequencing is an inherently chaotic sorting problem. Components arrive loose and unsorted in large storage containers. They are rarely oriented perfectly; parts shift during transit, overlap, or sit at odd angles. A robot cannot rely on a fixed trajectory to grab them. It must look into the container, understand the spatial orientation of a part, determine the optimal grasp point, and pick it cleanly without snagging adjacent pieces.
Once retrieved, Figure 03 places the component into a precise slot on a sequencing trolley. When full, these trolleys are hooked up to automated tugger trains or smart transport vehicles and rolled out to the line, delivering parts to human assemblers exactly when and where they are needed.
This requires what roboticists call dynamic loco-manipulation. The machine cannot simply stand still and move its arms. It must step, shift its torso, and counter-balance its weight while reaching into deep bins or pulling a heavy, wheeled cart.
Re-engineering for the Long Shift
To execute these complex logistics workflows, Figure 03 received a comprehensive hardware and software overhaul designed for continuous industrial output:
- The Helix 02 Brain: The robot operates on Figure’s proprietary vision-language-action (VLA) neural network. Helix 02 processes raw pixels from the robot’s cameras and data from its tactile sensors, translating that input directly into physical movement commands 200 times per second. A secondary control layer manages balance and motor stabilization at 1,000 times per second.
- Tactile Fingertips: The machine features upgraded five-finger hands equipped with palm cameras and highly sensitive pressure sensors in the fingertips. These sensors can detect forces as subtle as 3 grams, allowing the robot to feel whether a thin metal component is secure or beginning to slip out of its grasp.
- Mechanical Reliability: The forearm was the top hardware failure point during the Figure 02 pilot due to dense packaging and the thermal strain of three degrees of freedom. Figure 03 completely re-architects the wrist electronics, connecting the motor controllers directly to the main computer. This eliminated internal distribution boards and fragile dynamic cabling, significantly lowering the mechanical failure rate.
- Continuous Availability: The robot features a 2.3 kWh battery packed into its torso, yielding a 5-hour operating time. Instead of requiring manual battery swaps, charging coils are embedded directly into the machine’s feet. When power runs low, it simply walks to a wireless charging platform and tops off while standing upright.
The Macro Industrial Race
BMW’s aggressive push into what it calls “Physical AI” is a calculated response to shifting macroeconomic realities and an intense global race for industrial efficiency.
According to data from the International Federation of Robotics (IFR), global competition is magnifying, with China leading raw industrial installations. China recently accounted for roughly 45% of all global robotic installations in the automotive sector. For Western automakers, investing in adaptive, general-purpose humanoids is a strategy to maintain manufacturing flexibility without completely re-engineering existing factory layouts designed around human proportions.
Projections from the Boston Consulting Group suggest that the global market for humanoid robots could scale dramatically by 2030, with estimates ranging from 1,000,000 to over 6,000,000 units operating annually, depending on how quickly production costs fall.
For BMW, the target is clear: integrate these machines into environments that are monotonous, ergonomically straining, or safety-critical, freeing human workers to focus on higher-value assembly tasks.
“Plant Spartanburg is the birthplace of humanoid robotics in BMW Manufacturing’s operational day-to-day activities,” said Ulrich Wieland, Vice President of Production Control and Logistics at BMW Manufacturing. “Having already successfully completed a pilot with Figure 02 in our body shop, we are now looking forward to deploying Figure 03 for a sequencing use case in logistics.”
As Figure 03 begins its work in Hall 52, the question surrounding humanoid robots is no longer whether they can function in the real world. The focus has officially shifted to how quickly they can be built to scale. For procurement teams, industrial managers, and buyers tracking this deployment, platforms like the Anton Robots humanoid robot directory have already begun cataloging specifications, pricing, and certified commercial suppliers—marking the exact juncture where experimental bipedal machines transitioned into standard factory procurement listings.