Developments & practice
How Digit transfers totes onto conveyors at GXO
GXO is expanding its use of Digit. We explain the warehouse task and what more than 100,000 tote movements actually demonstrate.
Logistics provider GXO plans to deploy the humanoid robot Digit at two more US sites. The expansion builds on its work in Flowery Branch, Georgia, where the companies say Digit has moved more than 100,000 totes in total. The robot handles a defined step in material flow, placing containers onto conveyors.
The announcement of 26 October 2026 names Statesville, North Carolina, and Lebanon, Tennessee, as the next sites. Deployment is expected to begin in early 2027, initially using Digit 4. A move to Digit 5 is planned once that generation becomes commercially available. This planned expansion must be distinguished from the existing operation in Georgia.
For assessing the project, the specific task matters more than the robot’s human-like shape. Digit works within an established logistics environment. It does not process an entire order; it performs a physical handover. The contents of a tote and what happens to them after the conveyor belong to other parts of the overall workflow.
A handover between existing systems
According to the companies, collaboration began with a trial in late 2023. A multi-year agreement for commercial operation followed in June 2024. The original GXO announcement identifies the facility as one operated for clothing brand SPANX and describes integration with existing robotics. The agreement uses a model called Robots-as-a-Service: the robotic capability is supplied as a service.
Several conditions must align at such a handover. A tote must be ready for collection, and the robot must be able to grasp it and move it safely to the intended transfer point. The conveyor must then accept it. This explains the function of the interface; the announcements do not provide complete technical documentation of the installation.
A simple thought experiment shows why the interface matters. Assume a robot could transfer one tote every 30 seconds. If a tote arrives only once a minute, usable output is at most 60 transfers per hour. Faster grasping would not increase the supply. Conversely, if the conveyor stopped, the handover would have to wait for it to become available.
These figures are illustrative and are not performance values for Digit. They show why one movement cycle says little about the complete material flow. It also matters how often the robot waits for another system and how operation resumes after an interruption. Integration needs to handle these states as well as a successful grasp.
In its information about deployment, Agility describes three stages. First, the workflow is tested in a controlled environment. Next comes validation at the customer’s site outside the production area. Only then does the program envisage a 90-day operational pilot measuring availability, throughput and reliability, among other factors. This is the manufacturer’s published general process, not a retrospectively verified schedule for every GXO project.
The sequence has a clear purpose. A grasping trial can establish that a tote can be handled in principle. At the customer’s site, local space constraints and interfaces enter the picture. Operation then establishes whether the workflow fits the upstream and downstream steps over an extended period. Each stage therefore answers a different question.
What 100,000 totes tell us
According to the companies, the published total documents repeated tote movements in a working facility. It is neither an hourly throughput figure nor an availability percentage. The announcement provides no corresponding breakdown of robot numbers, productive hours, waiting time or human interventions. Dividing the total by the time since the contract began would therefore not yield a reliable performance measure.
Another illustrative calculation shows the difference. If 100,000 transfers are spread across 1,000 productive hours, the result is 100 transfers per hour. Across 2,000 hours, it is 50. Both scenarios produce the same total. Without the relevant operating hours, it is impossible to determine which scale applies. A different fleet size would also change the output per robot.
The announcement does not disclose the cost per transfer either. A business calculation would need to include, for example, internal support time and integration costs alongside the service fee. It would also need to establish which costs the contract already covers. The term Robots-as-a-Service alone describes neither a complete cost allocation nor a specific saving.
The decision to add two sites is nevertheless a concrete development. It shows that GXO and Agility intend to continue their collaboration and extend it to further operating environments. Whether the same results are achieved there will have to be established after the announced start. New sites may have different routes, staging arrangements and handover requirements.
The current position is therefore clearly bounded: in Flowery Branch, Digit’s tote handling forms part of an operating installation, and the companies report a six-figure total of completed movements. Two further deployments have been announced. A robust comparison with an alternative technical solution would additionally require measured hourly output, availability and total cost. This is news about an ongoing deployment, rather than a general economic case for humanoid robotics.
Sources checked: 12 November 2026. Project information comes from GXO and Agility Robotics; the calculations are fictional examples. Deployments announced for 2027 are not completed projects.