Humanoid or AMR?

Sort tasks by what the robot must do

Write down each task you want to automate and mark what it really involves: carrying a load from one place to another, picking an item, handling something awkward, or working at fixtures built for a person. Moving totes, cartons and pallets between known points is transport, and a wheeled mobile robot does it without arms or balance to manage. Picking and placing at one spot suits a fixed arm or a goods-to-person station. A humanoid's case rests on tasks that combine walking with two-handed handling in spaces you cannot easily change, such as shelving at varied heights or carts designed for staff. Tasks that fall into none of these groups may not be ready for any robot.

Weigh proven returns against early trials

Look at what large operators do with their own money. An A3 article from September 2026 reports that GXO planned nearly 20,000 robots across its warehouses that year and that none would be humanoids, even after 45 humanoid pilots. Its chief executive named hand dexterity and task speed as the gaps, and called humanoid picking of lipstick painfully slow. The same article says the established robot types already generate returns and can be deployed at scale. Read that as a sequencing lesson: deploy what pays today for the bulk of your moves, and keep a humanoid trial running alongside on a task where its human-like form genuinely matters.

Understand how mobile fleets make decisions

Autonomous mobile robots differ from older guided vehicles in where decisions are made. A 2024 book chapter by Fragapane, de Koster and Sgarbossa describes how their hardware and control software allow decentralised decisions and autonomous operation in changing surroundings, which has altered traditional planning and control methods. In practice, an AMR fleet can absorb some layout changes and congestion without new floor markers, but you still have to set rules for task allocation, charging and traffic. Ask any humanoid maker the same questions: how its robot receives work, shares space with your existing fleet and reports its status, so that you do not end up supervising two disconnected systems.

Match the choice to how often things change

Fixed automation such as conveyors, sorters and storage grids suits steady volumes and product sizes, but moving it later is costly. Mobile robots give up some of that density in exchange for the ability to reroute when a layout changes. A humanoid gives up more again, in return for working at stations built for staff without rebuilding them. So ask how often your layout, product range and customer mix are likely to change over the life of the equipment. A site with stable flows and a long lease points one way, while a short contract for a client whose range shifts each season points another. Record the answer beside each task on your list.

Sources and further reading

• GXO Plans 20,000 Robots in 2026. None of Them Will Be Humanoids. — Association for Advancing Automation (A3).
  Why one large logistics operator is scaling proven robot types while keeping humanoids in pilots, in its chief executive's words.
• Autonomous Mobile Robots for Material Handling in Intralogistics (book chapter, 2024) — Erasmus University Rotterdam (Research Portal).
  Explains how AMR hardware and software allow decentralised decisions, and what that changes for planning and control.

Common questions

Can a humanoid robot replace an AMR for moving totes?

It can carry totes, but carrying is the job a wheeled mobile robot already does with fewer moving parts and a simpler safety case. A humanoid makes more sense when the same trip also needs hands, for example loading shelves or carts that were designed for people.

Are any warehouses using humanoids in full production?

Reports so far point mainly to pilots. A September 2026 A3 report on GXO described dozens of humanoid trials running alongside a large fleet of other robot types, with production use of humanoids judged to be about two years away.