Signals Inbox·July 28, 2026·Humanoid Robotics

Is Figure already ahead in factory humanoids?

Figure has moved to the front of the technical race with a proven BMW deployment, a harder second assignment and a fast production ramp. Agility still runs the more mature commercial operation, while UBTECH leads on disclosed delivery volume.

Q1Why is Figure being called the factory humanoid leader now?

Figure looks like a factory humanoid leader today because its robots have moved beyond short pilots and returned to BMW with a harder job.

The first BMW project gave Figure something most competitors still lack: a completed deployment that BMW itself described as successful. Figure 02 worked in the body shop at BMW’s Spartanburg plant, loading sheet-metal pieces before they were welded into BMW X3 vehicles.

BMW has now brought in Figure 03 for a different workflow. The robot sorts components that arrive in inconsistent positions, places them in the correct order and moves the loaded trolley through the work area. It needs more perception and adaptation than a repeated pick-and-place sequence.

Figure has also started producing Figure 03 in the hundreds. The company is advancing on three fronts at once: factory experience, robot intelligence and manufacturing capacity.

That combination explains the sudden confidence around Figure. Several competitors have impressive robots, large orders or more operating hours. Few have recently shown this much movement across the whole system.

Q2What does being “ahead” in factory humanoids actually mean?

Figure is ahead under some definitions today, while Agility Robotics and UBTECH lead under others.

We cannot judge this race with one number. A company could lead because its robot handles the hardest task, works the most hours, serves the most customers or ships the most units. Factories will eventually care about all of those, plus uptime, safety and cost.

Figure currently makes its strongest case around dexterity, learned autonomy and automotive work. Agility has accumulated far more operating experience across customer facilities. UBTECH has delivered larger batches of industrial humanoids, especially in China.

Tesla remains harder to place. It has enormous manufacturing resources, although it still publishes little measurable Optimus factory data.

For now, Figure leads the more advanced end of the technical contest. The commercial race is still split.

Factory humanoid leaders by category, July 2026

Q3Are factory humanoids already general-purpose workers?

Factory humanoids are currently useful specialists, rather than workers that can take over any job a person performs.

Figure 02 loaded three types of sheet metal. Digit mainly moves totes and containers. Apollo is being tested for kitting, sorting, inspection and parts delivery. UBTECH’s Walker robots have handled boxes, components and basic inspection work.

These are real jobs, but they belong to a fairly narrow group. Most involve moving a known object between known points in an environment prepared for the robot.

We have not yet seen a humanoid begin the day loading a machine, switch to quality inspection, repair a fault and finish the shift assembling several products. Humans move between those jobs with limited preparation. Current robots still need training, integration and testing for each workflow.

Figure’s broader ambition is credible because Helix can control different tasks through the same robot. Still, the company’s best factory proof comes from one tightly defined BMW process.

For the moment, “general-purpose” describes where the technology is going, not how it is normally used inside factories.

Q4Did Figure really work on a live BMW production line?

Figure really did work inside BMW production, rather than performing a carefully staged demonstration beside the line.

BMW confirmed that Figure 02 inserted sheet-metal parts into a welding fixture at its Spartanburg plant. Industrial arms then welded those pieces before they entered the wider vehicle-production process.

This was a limited role, but mistakes carried real consequences. A robot feeding an active line has to place the correct component accurately and quickly enough to avoid slowing the surrounding equipment.

The deployment continued for months. Figure said the robots operated on working days and followed ten-hour shifts. BMW later approved another project using the newer Figure 03 platform.

That second decision is important. BMW had already seen the failures, interventions and integration work required by Figure 02. It still chose to continue.

Figure was not assembling entire cars. It performed one upstream task inside a much larger production system. Even with that narrower description, the BMW project remains one of the strongest pieces of factory evidence in the humanoid industry.

Q5How impressive were Figure’s BMW factory numbers?

Figure’s BMW results were large enough to prove repeatability, although they were still small compared with normal automotive automation.

Figure reported more than 90,000 parts loaded during over 1,250 operating hours. Those components contributed to more than 30,000 BMW X3 vehicles.

Because each cycle involved three components, the vehicle and component totals fit closely together. That gives the disclosure more substance than a vague claim about “supporting production.”

The runtime equals roughly 125 ten-hour shifts. That is several months of full-shift equivalents, far beyond the few hours usually shown in robotics demonstrations.

The totals also imply an average of about 72 part placements per logged hour. That figure includes waiting, calibration, stoppages and time when the robot may not have been cycling, so it cannot be treated as its maximum speed.

Automotive plants already contain industrial robots that complete millions of cycles with extremely high uptime. Figure has not reached that standard. What it proved was narrower but still meaningful: an early humanoid could work consistently enough to become part of a real vehicle-production process.

Reported results from Figure 02 at BMW Spartanburg

Q6Did Figure prove it could meet BMW’s speed and reliability targets?

Figure proved that it could remain useful at BMW, but it has not published a complete scorecard for speed, accuracy and human interventions.

Figure said the full cycle needed to fit within 84 seconds. It aimed to place all three pieces correctly in more than 99% of cycles and eventually complete a shift without a person resetting the robot.

Those are serious factory targets. The robot also had to position the pieces within five millimetres and perform each placement in roughly two seconds.

What Figure released later were cumulative totals, not the final result for every target. We still do not know its median cycle time, number of resets, failed-placement rate or unplanned downtime.

The average calculated from total parts and logged hours works out slower than an 84-second cycle. That does not show Figure failed, since logged time can include waiting and setup. But it is exactly why the missing breakdown matters.

BMW’s decision to continue with Figure 03 strongly suggests that the first robot cleared the minimum bar for usefulness. It does not prove Figure 02 consistently matched an experienced worker or mature industrial machine.

Q7Is Figure 03 doing a harder factory job than Figure 02?

Figure 03 is now tackling a harder BMW task, although the project has not accumulated enough results to prove it can do it reliably.

Figure 02 worked with three known pieces presented in a structured setup. Figure 03 receives components inside larger containers where they may have shifted, rotated or become partly hidden.

The robot must identify each part, choose a workable grip and place it in the correct position inside a sequencing trolley. It also adjusts its feet and torso while reaching, then pulls the loaded trolley.

That combination is closer to the work humanoid robots are supposed to unlock. Traditional automation performs best when every object arrives in almost exactly the same place. Automotive sequencing contains more variation, particularly when factories produce many vehicle configurations on the same line.

BMW says this kind of logistics workflow occurs frequently in vehicle production. Success could therefore open more opportunities than Figure’s first sheet-metal station.

Still, Figure has only shown the early deployment and demonstration. It has released no total for hours, correctly sorted parts, completed trolleys or interventions.

The difficulty has clearly gone up. Now the robot has to repeat it thousands of times.

Q8Is Figure ahead in dexterous factory work?

Figure currently has the strongest public case for dexterous humanoid work inside a major factory.

Figure 03 combines tactile fingertips, palm cameras and five-fingered hands with control of the robot’s arms, torso and legs. During the BMW sequencing workflow, it handles thin components, changes its body position and pulls a heavy trolley.

The hands are particularly important. Figure says the fingertips can detect very small forces and adjust when an object begins to slip. Palm cameras preserve a view of the object when the robot’s head cannot see around its arms.

Agility made a different choice with Digit. Its grippers are less human-like but well suited to moving totes. That can be more dependable for one job, even though it limits the range of objects the robot can naturally handle.

Apptronik and UBTECH also show robots using dexterous hands. Neither has published a customer case combining precise manipulation, locomotion and whole-body force with the same level of factory detail.

Figure’s lead here looks real. Whether those hands remain accurate after months of impacts, dust, vibration and constant use is another question. A sophisticated hand that spends too much time being repaired loses its charm pretty quickly.

Q9Does Figure have the best factory humanoid AI?

Figure currently shows the most convincing generalist humanoid AI, but factory results are still too limited to crown Helix as the best system.

Helix controls perception, movement and decision-making through one connected model. Figure 03 can adjust its hands, arms, torso and feet as the scene changes, rather than following only a fixed sequence of programmed positions.

That is visible in the new BMW workflow. When a component has shifted or the container is slightly misaligned, the robot can make small corrections while it moves.

Figure has also shown Helix completing longer household and logistics tasks. Those demonstrations suggest that the same intelligence can transfer across environments instead of being rebuilt for every motion.

The weakness is independent measurement. Most Helix demonstrations are produced by Figure, under conditions chosen by Figure. BMW has confirmed the factory project, but it has not released a detailed comparison between Helix and competing control systems.

Agility’s AI looks less dramatic in demonstrations, yet its robots have accumulated far more field time. A narrower system that works for thousands of customer hours may offer more practical value than a flexible model with less operating history.

Figure is ahead in visible AI capability today. There is not enough evidence to say Helix delivers the industry’s best factory uptime or productivity.

Q10Has Figure actually solved humanoid mass production?

Figure has built a serious robot-production line, although it has not yet shown sustained output at its advertised capacity.

Figure says BotQ has delivered more than 350 Figure 03 robots. It also reduced the demonstrated final production cycle from one robot per day to one per hour in less than four months.

The facility now uses more than 150 connected workstations and over 50 inspection points. Figure has separately produced more than 9,000 actuators and over 500 battery packs.

BotQ is plainly more than a workshop where engineers assemble a few prototypes by hand.

Quality remains a work in progress. Figure reported an end-of-line first-pass yield above 80%, meaning a meaningful share of completed robots still required additional work before passing. The battery line was much more mature, with a 99.3% first-pass yield.

The one-hour cycle is a production milestone, not proof that Figure makes one finished robot during every hour of the year. Even that continuous rate would equal 8,760 robots annually, below BotQ’s stated 12,000-unit capacity.

The 350 robots should not be mistaken for 350 robots already earning money at customer factories either. Figure has not said how many are used for internal testing, AI training, durability work or external deployment.

BotQ gives Figure one of the strongest manufacturing stories in the sector. Full mass production will require thousands of units, high yields and regular customer deliveries.

Q11Does Figure have enough factory customers to claim the lead?

Figure does not yet have enough disclosed customers to call itself the broad commercial leader.

BMW is a high-quality reference because the relationship continued after a long first deployment. A returning customer generally tells us more than another new pilot announcement.

Figure has also signed an agreement with Catalyst Brands for logistics and distribution work. The initial focus is a Nevada facility, with possible applications across the company’s wider network of retail brands.

Beyond those relationships, Figure has disclosed little. It may have customers operating privately, but we cannot count factories, robots or contracts that have never been named.

Agility works with GXO, Schaeffler, Toyota Motor Manufacturing Canada and Mercado Libre. UBTECH has announced industrial relationships across several Chinese carmakers and electronics manufacturers. Apptronik has ties to Mercedes-Benz, Jabil and GXO.

Figure may have the deepest single automotive case. Other companies have spread their robots across more organisations.

Today, Figure has one excellent factory reference and a second commercial route opening. Strong start, yes. Market-wide customer leadership, no.

Q12Is Agility Robotics commercially ahead of Figure?

Agility Robotics is ahead of Figure today in customer hours, contracted demand and breadth of commercial deployment.

Digit has accumulated more than 65,000 operating hours across commitments covering nine customer facilities. At GXO, it has moved more than 100,000 totes during a commercial deployment.

Agility also has active agreements involving Schaeffler, Toyota Motor Manufacturing Canada and Mercado Libre. Its latest company disclosure says customers have placed more than $300 million of multi-year orders for Digit v5, subject to contractual milestones.

Those numbers carry a different weight from pilot announcements. They show repeated customer work, a wider support operation and companies willing to make multi-year commitments.

Figure’s robot appears more dexterous, and the BMW sequencing task is more technically ambitious than moving standardised totes. A procurement manager may still prefer the supplier with years of operating data and several customers already under contract.

Agility has gaps of its own. Its latest Digit generation is entering commercial launch, and the company has not shown Figure’s pace of hardware production. Much of Digit’s proven work also remains concentrated around container movement.

Even so, Agility currently has the clearest commercial lead among American humanoid companies. Figure needs several more BMW-scale deployments to take that position.

Q13Is UBTECH already ahead of Figure in industrial humanoids?

UBTECH is ahead in disclosed industrial delivery volume, but Figure has published stronger proof of performance at one demanding factory.

UBTECH began mass production and delivery of Walker S2 with an initial batch of several hundred units. It also reported more than RMB800 million in Walker-series orders, equal to roughly $110 million.

The company has relationships with BYD, Geely, FAW-Volkswagen, Dongfeng, Foxconn and other manufacturers. Walker robots have been shown moving materials, sorting parts, performing inspections and supporting automotive processes.

This gives UBTECH a broader industrial network than Figure and a larger disclosed delivery programme.

The difficulty is separating delivered robots from productive robots. Some UBTECH orders involve data-collection centres or phased deployments. The company has not published a clear equivalent to Figure’s BMW totals showing customer runtime, successful cycles and production output for the wider fleet.

China is pushing humanoid production faster than the United States, helped by large supply chains and government-supported industrial programmes. UBTECH is one of the clearest examples.

UBTECH gets the volume lead. Figure has the stronger case for a carefully measured automotive deployment. They are different achievements.

Q14Which companies could overtake Figure next?

Tesla, Apptronik and Boston Dynamics could all overtake Figure, although none currently matches its combination of BMW evidence, dexterity and production progress.

Tesla has the most obvious scale advantage. It already operates large factories and controls much of its engineering, battery, electronics and manufacturing supply chain. It can test Optimus internally without waiting for an outside customer.

The missing piece is measurable factory evidence. Tesla has shown Optimus performing simple tasks and continues to prepare for larger production, but it has not disclosed customer hours, completed cycles, uptime or cost savings comparable with Figure or Agility.

Apptronik has raised more than $935 million in its Series A and works with Mercedes-Benz, Jabil, GXO and Google DeepMind. Jabil could become particularly valuable because it can manufacture Apollo and test it across real production environments. Apptronik still publishes few hard numbers about Apollo’s daily factory output.

Boston Dynamics brings decades of robotics work and access to Hyundai factories. Its electric Atlas looks highly capable, yet its planned factory deployments remain less mature than Figure’s completed BMW programme.

Any of the three could move ahead quickly. Tesla has manufacturing power, Apptronik has an unusually strong partner network and Boston Dynamics has deep robotics experience. Figure leads them now because it connected advanced technology to a measurable customer deployment sooner.

Q15Is Figure ahead on factory safety?

Figure has made sensible safety improvements, but nobody has proved that general-purpose humanoids can freely share busy factory spaces with people at scale.

Figure 03 is lighter than its predecessor and uses softer external materials around parts of the body. The design also reduces exposed pinch points, while the battery has passed recognised transport-safety testing.

Those features lower specific risks. They do not amount to complete certification for close, unsupervised work beside employees.

A factory safety case must cover falls, unexpected movements, collision forces, software faults, emergency stopping and what happens when a person enters the robot’s path. The answer changes depending on whether the robot is carrying an empty box or a heavy metal component.

Agility is making cooperative safety a central feature of Digit v5. The company says the robot is designed to become the first AI-enabled humanoid certified for this kind of shared work. “Designed to become” is still different from certification already completed.

Current humanoids often operate in controlled zones or carefully managed workflows. Figure appears to be taking safety seriously, but it has no clear lead based on public certification.

Q16Have Figure or its rivals proved that factory humanoids save money?

No humanoid company has published enough financial data to prove a reliable factory payback.

Figure and BMW have not disclosed the robot’s price, integration cost, maintenance expense, supervision needs or cost per correctly loaded component. We also do not know how Figure compares financially with a worker, a fixed robotic arm or a redesigned production cell.

The second BMW project suggests that the first deployment created enough value to continue. Encouraging, certainly. A renewed development project does not automatically mean the first robot was cheaper than the alternatives.

Agility is further along commercially. GXO has kept Digit in operation, and several companies have entered multi-year agreements. Its order book shows that customers are willing to reserve meaningful budgets for the next robot generation.

Even there, we lack a complete payback calculation. Moving 100,000 totes sounds substantial, but the economics depend on how many robots were used, how often people intervened and what the customer paid.

Humanoids may first make sense where staff turnover is high, injuries are common or traditional automation requires an expensive building redesign. They do not need to beat human wages in every situation.

Today, factories are paying for a mixture of useful work, learning and future optionality. The industry has not proved that humanoids are broadly the cheapest solution.

Q17Is Figure learning faster than its competitors?

Figure currently appears to be improving faster than most rivals, and that may matter more than leading every metric today.

The company moved through three robot generations quickly, completed an extended BMW deployment and retired the entire Figure 02 fleet so its lessons could be built into Figure 03.

This was more than a cosmetic redesign. At BMW, the forearm became the largest hardware failure point. Figure then simplified the wrist electronics, removed dynamic cabling and reduced the number of components that could fail.

At the same time, Helix moved from controlling mainly upper-body actions to coordinating the full robot. BotQ increased its demonstrated production speed by 24 times.

Most robotics companies improve hardware, AI and manufacturing on different schedules. Figure is pushing all three together. Factory failures change the next robot, a larger fleet creates more training data, and better software gives customers more possible tasks.

Agility has a much larger collection of customer operating data. UBTECH is learning from more delivered robots. Figure’s advantage lies in how quickly it turns each lesson into a new integrated platform.

If Figure 03 performs well at BMW and several new customers receive production robots, its technical lead could become a wider market lead. If those deployments stall, Agility’s slower and more proven approach may look wiser.

Q18Is Figure already ahead in factory humanoids?

Yes, Figure is already ahead in the most advanced part of the factory humanoid race, but it has not taken the overall commercial lead.

Figure produced one of the best-documented automotive deployments in the industry. Its new robot is now attempting a harder logistics workflow that combines perception, dexterous manipulation, walking and forceful cart movement.

Helix and Figure 03 also give the company a strong technical edge. No rival has publicly connected tactile hands, whole-body learned control, an extended BMW project and a production fleet quite as convincingly.

Agility remains ahead where commercial buyers may care most today. It has far more customer operating hours, more active enterprise relationships and over $300 million in conditional multi-year orders.

UBTECH has shipped more industrial humanoids, while Tesla could eventually manufacture at a scale that changes the entire ranking. None of these companies has published a complete factory ROI case.

So, is Figure already ahead? For difficult automotive tasks, dexterity and development speed, yes. Across customers, working hours and commercial orders, no.

Figure currently builds the most convincing advanced factory humanoid, while Agility runs the more mature humanoid business. Figure has moved to the front of the technology race. It still has work to do before it can claim the whole market.

Overall factory humanoid position, July 2026