In January 2024, Figure AI announced a partnership with BMW Manufacturing to deploy humanoid robots at BMW's Spartanburg, South Carolina plant — the largest BMW manufacturing facility in the world by volume, producing roughly 1,500 vehicles per day. The announcement attracted significant attention, partly because of Figure's profile as one of the better-funded humanoid startups, and partly because automotive assembly represents a meaningfully different challenge from the warehouse environments where most humanoid pilots have concentrated.
More than a year on, the deployment has proceeded quietly — which is to say, without the detailed public reporting that would let anyone outside the programme evaluate what's actually happening. What we can do is examine what the automotive manufacturing environment actually demands of a humanoid robot, and use that as a lens for understanding how close — or how far — the current generation of systems is from performing it reliably.
The answer is more nuanced than either the optimistic press coverage or the reflexive scepticism suggests.
Why Automotive Assembly Is a Different Problem
Warehouse environments — the setting for most humanoid pilots to date — have properties that make them relatively tractable for early-stage systems. Tasks are repetitive. Objects are standardised. The consequences of a failure are usually modest. And the environment, while busy, is largely designed around logistics rather than precision.
Automotive assembly plants share some of those properties but add several that are substantially harder. The tasks involve much tighter tolerances — fitting a component into a designated location with millimetre-level precision, rather than moving a bin between two points. Many assembly operations require coordinated two-handed manipulation, where both arms must work in sequence or simultaneously. Components vary in weight, shape, and handling requirements across different vehicle configurations. And the consequences of error are higher: a misaligned part that passes undetected can create quality or safety problems downstream.
BMW's Spartanburg plant builds the X3, X4, X5, X6, X7, XM, and M versions of several of those models — multiple vehicle lines with different configurations running on the same production lines. That variability is exactly the kind of condition that makes automation harder. Fixed industrial robots, which have operated in automotive plants for decades, handle it through careful programming for each specific task and configuration. A humanoid robot, by contrast, is supposed to handle variability more flexibly — but current systems' ability to do that reliably in production conditions remains genuinely uncertain.
What Figure Has Actually Demonstrated
Figure published a demonstration video in March 2024 showing its Figure 01 robot performing a task at a BMW facility — specifically, placing sheet metal body parts into fixtures on a production line. The video showed the robot identifying parts from a bin, picking them up, and placing them in the correct position in a fixture. The task was performed slowly and deliberately, and the video was edited — meaning we see successful completions, not the full range of attempts including failures.
That caveat matters, but it doesn't invalidate what the video shows. The manipulation involved — picking irregular metal parts from a bin and placing them accurately into a fixture — is genuinely more complex than moving uniform tote bins. The fact that Figure could demonstrate it at all, in a real manufacturing environment rather than a lab, is a meaningful data point about the state of the technology.
What the video cannot tell us: the success rate across many repetitions, the cycle time relative to human workers doing the same task, how the system handles parts that are slightly misaligned in the bin or fixtures that aren't perfectly positioned, and how much human oversight was present during filming. These are the questions that separate a credible demonstration from a production-viable system, and none of them have been answered publicly.
Figure announced in February 2024 that it had raised $675 million at a $2.6 billion valuation, with investors including Microsoft, OpenAI, Nvidia, and Jeff Bezos. That capital position gives the company substantial runway to continue developing its systems toward production viability. It does not, by itself, tell us whether the BMW deployment is performing as hoped.
The Skills Gap That Automotive Exposes
Watching Figure's demonstration video carefully, and comparing it to how a human worker performs the same kind of task, reveals something instructive about where the current generation of humanoid systems sits technically.
Human workers in automotive plants do not think consciously about how to pick up a part. They reach, grasp, adjust grip based on real-time feedback from their hands, and place — the entire sequence happening fluidly and quickly, drawing on years of embodied experience. When something is slightly off — the part is at an odd angle, the fixture has a burr on it, the lighting is different — they adapt without breaking stride.
Current humanoid robots handle the nominal case reasonably well. The part is where it's expected to be, the fixture is clean, the lighting is consistent. Under those conditions, modern manipulation systems can perform tasks that looked impossible five years ago. What they handle poorly is the non-nominal case — the situation that differs slightly from what the system was trained or programmed to expect.
In a BMW plant running 1,500 vehicles per day, the non-nominal case happens constantly. Parts shift in bins during transport. Fixtures wear and develop slight dimensional variations. Workers bump things. This is not an exotic edge case — it is the ordinary texture of a production environment running at volume. Any humanoid system that needs to operate reliably in that context needs to handle these variations gracefully, and the honest assessment of current systems is that they do not yet do this consistently.
What a Realistic Deployment Scope Looks Like
The most credible near-term deployment scenario for humanoids in automotive manufacturing is not general-purpose assembly work. It is a specific category of tasks that BMW and other manufacturers have already identified as problematic: ergonomically demanding jobs that cause high rates of injury among human workers, performed in locations where fixed automation is impractical.
Loading parts into overhead fixtures, working in confined spaces like wheel wells and engine bays, performing tasks that require reaching in awkward positions for extended periods — these are jobs that automotive manufacturers actively want to automate, have struggled to address with conventional fixed robots due to space and variability constraints, and that a capable humanoid could, in principle, perform without those constraints.
BMW has been explicit that this is part of the rationale for the Figure partnership. The framing is ergonomics and worker safety as much as pure productivity. That framing is worth taking seriously, because it defines a deployment scope that is more achievable than general assembly work and has a clear value proposition that doesn't depend on the humanoid matching human productivity from day one.
The timeline for even this constrained deployment scope reaching reliable production operation — rather than supervised pilot — is not publicly known. Figure has not announced when it expects to have systems running unsupervised shifts at Spartanburg. BMW has not commented on the programme's progress in specific terms. What is visible publicly is that the pilot is ongoing and that neither party has indicated it has failed.
The Broader Significance of the Automotive Test
Automotive manufacturing matters as a test environment for humanoid robotics for reasons beyond the specific companies involved. It is one of the most demanding production environments in existence — high volume, high precision, high variability, high consequence of error — and it has been the proving ground for industrial automation for half a century. If humanoid robots can operate reliably in automotive assembly, they can operate reliably almost anywhere.
The Figure-BMW partnership, whatever its current status, is the most serious attempt yet to find out whether that threshold is within reach. The fact that it is happening at all — that a major automotive manufacturer decided the technology was worth evaluating in a live production environment — tells us something about where the industry's expectations have shifted over the past few years.
Whether the technology is ready to meet those expectations is a different question, and one that will be answered by production data rather than demonstration videos. That data will emerge eventually, in BMW's quality metrics and production reports and, eventually, in what Figure chooses to say publicly about what it has learned. The gap between what gets announced and what gets verified is where the real story of humanoid deployment is currently being written.