---
service: "Publicasta"
schema_version: "1.0"
article_id: 978
title: "GXO’s Digit Expansion Tests Whether Humanoid Robots Can Become a Repeatable Warehouse Service"
language: "en"
default_language: "en"
canonical_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=en"
json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=en"
api_url: "https://publicasta.com/api/public/v1/channels/robots/articles/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=en"
channel_url: "https://publicasta.com/api/public/v1/channels/robots"
channel_articles: "https://publicasta.com/api/public/v1/channels/robots/articles"
search_url: "https://publicasta.com/api/public/v1/search"
documentation_url: "https://publicasta.com/api-docs#reading-publicasta"
openapi_url: "https://publicasta.com/api-docs/openapi.json"
published_at: "2026-10-26T17:08:10+00:00"
updated_at: "2026-10-26T17:08:10+00:00"
translations:
  - language: "ar"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=ar"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=ar"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=ar"
  - language: "de"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=de"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=de"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=de"
  - language: "en"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=en"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=en"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=en"
  - language: "es"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=es"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=es"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=es"
  - language: "fr"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=fr"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=fr"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=fr"
  - language: "pl"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=pl"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=pl"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=pl"
  - language: "ru"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=ru"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=ru"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=ru"
  - language: "zh"
    html_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026?lang=zh"
    markdown_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.md?lang=zh"
    json_url: "https://publicasta.com/robots/gxo_digit_multisite_humanoid_deployment_operating_case_2026.json?lang=zh"
---

# GXO’s Digit Expansion Tests Whether Humanoid Robots Can Become a Repeatable Warehouse Service

> GXO plans to add Digit robots at two U.S. sites after reporting more than 100,000 totes moved in Georgia. The meaningful test now is not whether a humanoid can work, but whether the deployment model can be repeated without hiding its operational limits.

GXO and Agility Robotics have announced the next practical test for humanoid robots in logistics: moving from one live warehouse to several. The companies say Digit robots will expand to GXO facilities in Statesville, North Carolina, and Lebanon, Tennessee, with deployments expected to begin in early 2027. The machines will initially be Digit 4 systems and are expected to migrate to Digit 5 when that model becomes commercially available.

 ![Humanoid warehouse robot moving a tote beside a conveyor in an active fulfillment center](https://publicasta.com/storage/projects/11/pages/978/2026/10/de788b12-6a69-4029-ba8a-0832cbfccc03.webp)

 That sounds like another deployment announcement. It is more useful to read it as an operating-model test. The question is no longer simply whether a bipedal robot can pick up a tote and place it on a conveyor. GXO and Agility now have to show that the same kind of work can be introduced at different sites, connected to different local processes, supported by a service organization, and measured against warehouse economics.

 The expansion follows the companies’ 2024 multi-year Robots-as-a-Service agreement at GXO’s Flowery Branch, Georgia, facility. According to the companies, Digit has moved more than 100,000 totes there. Agility separately says Digit has accumulated more than 65,000 hours of real-world operation across commercial operations and customer programs. Those are company-reported figures, not an independent audit of uptime, labor savings, injury rates or total cost per tote. They are still significant because they describe repeated work inside an active fulfillment environment rather than a controlled demonstration.

 ## What is actually changing

 The new facilities are not a promise that Digit will take over an entire warehouse. GXO says the robots will support e-commerce fulfillment by performing highly repetitive material-handling tasks. The earlier Flowery Branch deployment placed totes onto conveyors while working alongside existing automation. That narrow assignment is important. Warehouses contain many tasks that are repetitive, but they are not equally suitable for automation. A tote-handling loop with predictable handoff points is a more realistic starting point than mixed-item picking, exception handling or work that depends on subtle packaging judgments.

 The initial rollouts will use Digit 4. That matters because the announcement is partly about the current system and partly about a future product transition. Digit 5 was unveiled in September as Agility’s next-generation humanoid. The company describes it as designed for “cooperatively safe” work near people, with a new safety architecture, a higher payload and faster charging. But the new GXO sites are expected to begin with Digit 4, so the first operational question will concern a system that already has deployment history, not the full set of specifications Agility is presenting for the next generation.

 This staged approach may be sensible. A warehouse operator does not need to wait for a new model to begin learning how the robot interacts with work instructions, conveyors, shift changes, maintenance routines and human supervisors. At the same time, migrating from one generation to another creates its own work. A new robot can change reach, payload, charging schedules, behaviors around people and the requirements for spare parts or technician training. “Migration” is not automatically a plug-in replacement.

 ## Why the site count matters more than the robot count

 A single successful installation can be explained by unusually strong local conditions: a prepared work cell, experienced staff, supportive engineering teams and a task selected after months of iteration. A multi-site program tests whether those conditions can be recreated.

 Statesville and Lebanon will have their own floor layouts, inventory patterns, staffing arrangements and integration details. Even if both facilities use the same warehouse software, the physical work may differ. A conveyor can be positioned slightly differently. Tote presentation can vary. Lighting, floor markings, pedestrian traffic, rack geometry and congestion can affect how a mobile manipulator performs. A robot that looks general-purpose at the product level may still need a highly specific workflow at the facility level.

 That is why the most consequential part of the announcement is not the humanoid form. It is the combination of a customer with a large operational footprint and a supplier that is trying to standardize deployment. GXO operates more than 1,000 facilities, according to its company profile. If Digit becomes useful in only one specially tuned process, its addressable market is limited. If the partners can define a repeatable method for selecting tasks, preparing sites, integrating software and handling failures, the same hardware could become a service that is easier to evaluate.

 The distinction is familiar from other industrial automation projects. Equipment rarely scales because a machine worked once. It scales when the organization can install, commission, monitor, repair and improve it without treating every new site as a custom research project.

 ## Robots-as-a-Service changes the buying question

 The GXO agreement is structured as Robots-as-a-Service rather than a conventional equipment purchase. The public announcement does not disclose the commercial price, guaranteed uptime, payment formula or the precise allocation of maintenance costs. That missing information limits what can be concluded about return on investment.

 The model nevertheless changes the buyer’s exposure. A warehouse operator that buys a robot must usually carry more of the capital risk and may own a larger share of integration and maintenance responsibility. A service arrangement can shift some of that burden to the vendor, but it also makes performance clauses, response times, data access, software updates and exit terms central to the contract.

 For a logistics operator, the useful calculation is not the robot’s list price. It is the cost of completing the target workflow over time. That calculation should include:

 - productive operating hours rather than powered-on hours;
- throughput under normal and disrupted conditions;
- human supervision and exception-handling time;
- charging, battery replacement and planned maintenance;
- integration with warehouse-management and conveyor systems;
- training for operators, technicians and safety staff;
- downtime while a vendor diagnoses a remote or software-related fault;
- the cost of changing the process if the robot cannot handle a new packaging or inventory pattern.

 A RaaS contract can make the first deployment easier to approve, but it does not make the economics automatically attractive. A service fee is still a cost per operation. The operator needs enough visibility to compare it with labor, conventional automation and the cost of leaving the task unchanged.

 ## The 100,000-tote milestone is useful, but incomplete

 Moving 100,000 totes is a meaningful endurance signal. It suggests that Digit has been used repeatedly in an active facility and that the deployment has survived beyond a short publicity demonstration. It does not, by itself, show that the robot is cheaper than alternatives or that it performed without substantial human intervention.

 The missing metrics are the ones procurement and operations teams will eventually need. How many totes were moved per hour? What percentage of cycles required a person to intervene? How did performance change across shifts? How often did the robot pause because of a perception problem, a blocked route, a conveyor issue or a software fault? What was the mean time to recover? How much floor space was reserved for the system, and how often did that space limit other work?

 There is also a difference between a total milestone and a steady operating rate. A robot can accumulate many completed cycles while spending considerable time in engineering support or operating only during carefully chosen periods. The announcement does not provide a public denominator for the 100,000-tote figure. Readers should therefore treat it as evidence of sustained use, not as a complete productivity benchmark.

 Agility’s own description of Digit’s deployment process points toward the right kind of evidence: fleet visibility, throughput, uptime and Mean Time Between Incidents through its Arc platform. Those measures are more informative when reported with definitions and time windows. “Uptime” can mean that the robot is available, powered, connected or actively completing work. Those are different states. A credible deployment report should explain which one is being measured.

 ## Digit 5 could remove some barriers, but its claims are still forward-looking

 Agility says Digit 5 can repeatedly lift loads up to 50 pounds, reach up to 7.2 feet and operate with a 10:1 run-to-charge ratio. The company describes a 90-minute runtime battery that can charge in nine minutes, with more than 20 hours of productive work possible across a 24-hour day. These are preliminary product claims based on internal testing and pre-production configurations, and Agility’s product materials say specifications and safety features may change.

 Those figures are relevant to warehouse design, but they should not be confused with a guaranteed shift schedule. A robot’s useful day includes travel, waiting for a handoff, clearing an exception, responding to a person in its path, charging, inspections and maintenance. A 10:1 run-to-charge ratio is a battery and charging claim. It is not the same as a 10:1 ratio of productive work to total elapsed time.

 The safety architecture is another important development. Agility says Digit 5 uses multiple sensor technologies and proprietary algorithms to detect people, then avoid, stop or assume a seated position. It also describes visual and auditory safety cues and an independent safety controller. The stated goal is to reduce reliance on physical barriers in areas where people and robots work close together.

 That could make a large difference to deployment cost and floor-plan flexibility. Traditional automation often requires fences, interlocks or dedicated cells. A robot that can safely share space with people may fit into processes that would be too expensive or disruptive to isolate. But “designed for cooperatively safe work” is not the same as universal permission to operate in every mixed-traffic environment. The site still needs a risk assessment, defined operating zones, emergency procedures, training and validation of the complete system, including conveyors and nearby equipment.

 Safety also has to be evaluated over the failure modes that matter in production. A robot should not only stop when a person enters its path. Operators need to know what happens after a dropped tote, a blocked conveyor, a misidentified object, a loss of network connectivity, a sensor obstruction or an unexpected change in the route. Agility has said that core safety functions are onboard and separate from ordinary cloud connectivity, which is a useful architectural distinction. It does not remove the need to test the robot in the actual facility.

 ## The humanoid shape is not the main economic argument

 GXO’s use case does not require Digit to imitate a person in every respect. The value proposition is that a human-scale robot can use parts of a facility that were built around human workers: aisles, shelves, handoff points and workstations. That flexibility may reduce the need to redesign a site.

 It can also create a trap. Human-compatible spaces are not necessarily robot-friendly spaces. People adjust constantly to clutter, changing traffic and ambiguous objects. A robot may need explicit rules for all of those conditions. If the facility must be reconfigured extensively, then part of the supposed advantage of the humanoid form is spent on preparation.

 The right comparison is therefore task-specific. If a fixed robotic cell can perform the same operation faster, more safely and at lower cost, a humanoid does not win merely because it can walk. If the task changes frequently, the work area is difficult to isolate or the facility cannot justify a bespoke automation line, a mobile humanoid may have a stronger case. The benefit is flexibility under operational constraints, not human likeness as an end in itself.

 For GXO, the business case may also include the ability to offer automation across different customers and buildings. A contract logistics provider manages sites for many brands, and those customers may have changing volumes or product mixes. A service-based, reconfigurable system could be valuable if it can move between workflows without a long installation project. Again, the key evidence will be the cost and time of reconfiguration, not the number of capabilities in a product brochure.

 ## What the new deployments should measure

 The Statesville and Lebanon launches create an opportunity to publish a more complete operational record. A useful scorecard would separate robot performance from the surrounding warehouse process.

 At the robot level, operators should report task success, intervention rate, cycle time, route interruptions, battery behavior, preventive-maintenance hours and unscheduled downtime. At the workflow level, they should measure end-to-end throughput, queueing at handoff points, the effect on conveyor utilization and the amount of work shifted to human staff. At the business level, they should compare cost per completed tote, service fees, labor redeployment, safety incidents and the time required to recover from a failure.

 The comparison group matters. A warehouse can improve after a robot arrives because managers also change layout, instructions or staffing. If all improvement is attributed to the robot, the result becomes difficult to reproduce. The cleanest evaluation would compare similar shifts or workflows before and after deployment, while clearly stating what else changed.

 The partners should also explain how they choose tasks. A robot that handles one tote type reliably may not handle another with a different center of gravity, label placement or packaging stiffness. The public should learn whether the new sites use the same tote and conveyor interfaces as Flowery Branch, whether the tasks are deliberately identical, and how many exceptions the robot is expected to manage.

 ## A practical reading for warehouse buyers

 For operators considering a humanoid pilot, the GXO announcement suggests a cautious sequence. Start with a repetitive workflow that has a clear input, output and handoff. Record the current process before introducing the robot. Define what counts as success and what human support is acceptable. Require a recovery plan for common failures. Put software, cybersecurity, data retention and remote-support rights in the contract. Make sure the service arrangement explains what happens if the vendor misses availability targets or the product roadmap changes.

 The pilot should also have a stopping rule. A demonstration can continue indefinitely if the goal is merely to produce more footage or collect more examples. An operational pilot needs a deadline and a decision framework: expand, redesign the task, keep the system at the current site or stop. That framework protects both sides from turning an interesting experiment into an unmeasured permanent expense.

 The same discipline applies to labor planning. A humanoid may remove lifting and repetitive walking from a role without eliminating the need for people. Supervisors may spend more time managing exceptions, technicians may need new skills, and workers may become responsible for coordinating a mixed human-machine area. A deployment that improves ergonomics but adds hidden supervisory work can still be worthwhile, but the trade-off should be visible.

 ## The real milestone is repeatability

 GXO’s expansion is stronger evidence than a product launch because it commits the partners to additional operating environments. Yet the announcement is not proof that humanoid warehouses have reached commodity economics, nor does it establish that Digit 5 is ready for unrestricted production use. The two new sites are scheduled for 2027, the initial robots will be Digit 4, and the public materials do not disclose prices or independent performance data.

 The most important result will be whether Agility and GXO can transfer the deployment method. Can they select a suitable task quickly? Can they install the system without redesigning the whole building? Can supervisors understand its state and recover from failures? Can the service model keep the robot productive when the facility changes? Can safety performance be demonstrated in the actual traffic pattern rather than inferred from a controlled test?

 If the answer is yes, the partnership will show something more useful than a humanoid performing a warehouse task. It will show a path for turning flexible robots into managed infrastructure. If the answer is only partly yes, that is still valuable information. The next phase of robotics will be judged less by whether a machine can perform a task once and more by whether an organization can depend on it every day, at several sites, with costs and limits that are visible to the people buying the service.