Humanoid Robots in Warehouses and Factories: How Much of the Job They Actually Do

The job, and why humanoids are aimed at it

O*NET, the US Labor Department's occupation database, describes this role as people who "manually move freight, stock, luggage, or other materials, or perform other general labor", under job titles such as loader, material handler, receiver and warehouse worker. The occupation employed 2,940,300 people in the United States in 2025, at a median wage of $19.35 an hour, and is projected to grow "slower than average (1% to 2%)" between 2025 and 2035, according to Bureau of Labor Statistics data shown on O*NET.¹ That projection isn't a forecast about robots.

A humanoid robot is a robot with a roughly human body plan, two arms and usually two legs. The pitch for using one in a warehouse is that it fits spaces built for people. When GXO began testing Agility Robotics' Digit in 2023, Agility's co-founder said "there's no need to redesign the warehouse or install infrastructure beyond that which is already designed around people."² GXO gave Digit's specification as 5'9" tall, 140 pounds, able to lift up to 35 pounds.² GXO has described humanoids that could move "from performing unloading functions on first shift, to picking on second, and then switch to packing", but only "when humanoids reach full operational maturity".¹⁷

The one production case: moving totes at GXO

GXO, which calls itself the world's largest pure-play contract logistics provider, announced a proof-of-concept pilot of Digit on December 6, 2023, at the SPANX facility in Flowery Branch, Georgia, which GXO runs for the womenswear brand. Digit was taught "repetitive tasks such as moving totes from autonomous mobile robots (AMRs) and placing them onto conveyors".² Autonomous mobile robots are wheeled carts that navigate a building on their own.

On June 27, 2024, GXO and Agility signed a multi-year robots-as-a-service agreement, under which GXO pays a subscription rather than buying the robots. GXO's chief automation officer said the company was "deploying fully operational Digit humanoids into a live warehouse environment", where the robots move "totes from cobots" onto conveyors, "all orchestrated through Agility Arc", Agility's fleet software.³ Two months later GXO wrote that it had "trialed three humanoid prototypes and became the first logistics provider to deploy this technology in a live operating facility", while adding that "there is a long development journey ahead to achieve mass deployment".¹⁷ Those operator statements are why the tote-moving task is graded as a production deployment.

The later numbers come from Agility, not GXO. In its September 30, 2026 SEC registration statement, filed for a planned stock listing through a merger with Churchill Capital Corp XI, Agility said Digit "has automated tote fulfillment and handling workflows" at GXO, "reaching a cumulative 100,000 tote milestone in November 2025 and maintaining approximately 98% operational accuracy while on-task".⁴ The filing doesn't define "on-task" or say how many robots moved those totes. GXO hasn't published a robot count either.

Pilots that ended, and pilots that converted

BMW ran the longest operator-confirmed humanoid job in a car plant. In August 2024 it reported a "trial run lasting several weeks" at Plant Spartanburg, South Carolina, in which Figure AI's Figure 02 "successfully inserted sheet metal parts into specific fixtures", and said that, at that point, "there are no Figure AI robots at BMW Group Plant Spartanburg, and there is no definite timetable established for bringing Figure robots to the plant."¹³ The robots came back. In June 2026 BMW said Figure 02 "supported the production of more than 30,000 BMW X3 vehicles over ten months" in 2025, loading sheet-metal parts for welding, and a BMW production vice president called it "a pilot with Figure 02 in our body shop".⁵

Figure's own account, published November 19, 2025, gives an "11 month" deployment, a "10-hour shift Monday-Friday", "90,000+ parts loaded" and "1,250+ hours of runtime". It lists the targets it set: an 84-second cycle with 37 seconds to load, more than 99% of cycles with all three parts placed correctly, and zero human interventions per shift. It doesn't report how close the robot came to those targets, and it says the forearm was its "top hardware failure point". The same post announced Figure 02's retirement and its "return to HQ from BMW".¹⁰ This work is closer to an assembler's or a machine loader's job than a hand mover's, so it informs the scorecard only indirectly.

Schaeffler, the German motion-technology supplier, describes Digit as "loading and unloading a washing machine" at its plant in Cheraw, South Carolina, "since the beginning of 2025", and calls it a "pilot project".¹¹ Schaeffler is also an investor: when it took a minority stake in Agility in November 2024, Agility's CEO said scaling humanoids would require them to work "outside of the safety cages they are in today".¹⁸ Agility's filing says Digit has moved "approximately 25,000 totes" at Schaeffler with "approximately 98% operational accuracy", and that Schaeffler "has committed to scaling its use of Digit 5, moving from pilot to full-scale integration".⁴ That commitment is Agility's statement; we found no Schaeffler source for it.

Toyota Motor Manufacturing Canada signed a subscription agreement in February 2026 "following a successful pilot", and its president said it would "deploy Digit to improve the team member experience and further increase operational efficiency".¹² Neither company has itemized Digit's tasks or robot count there.

Demonstrations and plans

BMW's next step is a logistics task. Figure 03 "will now start at Spartanburg", picking unsorted parts out of large containers and sorting them into a sequencing trolley, which an automated tugger train or transport robot then takes to the line.⁵ BMW's headline says Figure "demonstrates" the robot in the new use case, so it is graded as a prototype until BMW reports it running in live work.

Mercedes-Benz signed a pilot agreement with Apptronik in March 2024 to explore Apollo for "delivery of assembly kits", "inspecting the components" and moving "totes of kitted parts".⁷ A year later, heise reported that "fewer than ten" Apollos had been working at Berlin-Marienfelde and Kecskemét, Hungary, "deployed in a test environment of a production plant" and taught via teleoperation (remote control by a person) by experienced Mercedes-Benz workers, with more autonomy as the next step.⁸

Hyundai Motor Group said at CES in January 2026 that "Beginning in 2028, Atlas will be introduced on processes with proven safety and quality benefits, such as parts sequencing", extending to component assembly "By 2030".¹⁴ In September 2026 its subsidiary Boston Dynamics opened a training center inside Hyundai's Georgia plant, where Atlas robots are "being trained in real-world environments", and restated a group plan for "25,000 units at Hyundai Motor and Kia's global plants over the next few years". The same release still calls Atlas a humanoid platform "currently in development".⁶

Tesla's July 2026 quarterly update said it was "installing the first-generation lines for Optimus, where we expect to start production soon", and that "The initial Optimus builds will be used in our Optimus Academy for training data collection and further functionality development."¹⁵ In China, UBTech reported selling 1,079 full-size humanoids in 2025 and said Walker S2 targets "handling, sorting and quality inspection".⁹ Its results announcement names no operator, so those cases stay vendor claims here.

What it costs, and what has been measured

No operator in this post has disclosed what it pays. The most detailed figures come from Agility's filing, and they are illustrative assumptions for its next robot, Digit 5, whose prototypes were unveiled on September 15, 2026, with early access "expected to begin in the first half of 2027". Under the subscription model a robot generates "a recurring monthly fee of $8,500" plus a one-time deployment fee of about $25,000; a buyer would pay "approximately $200,000" upfront plus about $36,000 a year for software and maintenance.⁴ For comparison only, the occupation's 2025 median annual wage was $40,240.¹ The two aren't like for like: a robot's subscription covers hours, shifts and supervision that the filing doesn't spell out.

Agility reported total net sales of $1,781,967 in 2025, of which robot sales were $1,550,360, and two customers each accounted for 10% or more of revenue.⁴ It says it holds "$300 million of committed multi-year orders" for Digit 5, "subject to the satisfaction of certain contractual milestones", and its risk factors say one customer "has committed to deploy $300 million of Digit 5" over several years.⁴ These are orders, not robots at work.

The measured results are few and come from the makers: tote counts and an accuracy rate from Agility, parts loaded and runtime hours from Figure, and BMW's own vehicle count. None is independently audited, and none reports interventions per shift, robots per site or effects on staffing.

What the evidence does and doesn't show

Measured against the role's tasks, humanoids reach one in production: moving totes between machines and conveyors at one GXO site. Sorting, carrying kits and inspection have reached demonstrations or plant test areas, and the rest of the job (protecting stock, tagging, rigging, packing, reading work orders) has no humanoid case of any grade we could find. Agility says Digit is "deployed or committed for deployment across nine customer facilities", with more than 65,000 hours of commercial operation as of May 2026.⁴ The occupation employed 2,940,300 people in 2025.¹

None of the operators here has linked a humanoid to job cuts. The companies frame the robots as taking over repetitive or strenuous steps: GXO says they reduce strain and free team members "to do more value-added work",² and BMW points to "monotonous, ergonomically demanding, or safety-critical activities".⁵ Those are the companies' stated aims, not measured outcomes, and the step from one production case to an occupation of almost three million people is one no source in this post has measured.