Warehouse Picking Robots: What's Deployed at Scale in 2026

Amazon operates more than 750,000 robots across its global fulfilment network. Ocado's customer fulfilment centres run entirely on automated systems with no human pickers in the main grid. The largest third-party logistics operators — DHL, GXO, Geodis — have active robot deployments measured in thousands of units.

Warehouse picking robots are not emerging technology. They are deployed infrastructure. The question for logistics operators in 2026 is not whether to automate, but which system fits their operation, their building, their order profile, and their capital structure.

This article covers what is actually deployed at scale, how each system category works, and how to match the right system to the right operation. For the workforce implications, see Will Robots Replace Warehouse Workers? and the Warehouse Picker/Packer profile on the Geppetto Jobs Index.

This Is Not Emerging Technology

The framing matters. Industry coverage of warehouse robots frequently treats automation as a future development — something operators should begin planning for. That framing is approximately five years out of date.

Amazon's 750,000+ robot figure represents the largest civilian robotics deployment in history. It includes Kiva-derived shelf-moving AMRs (now Amazon Robotics), Sequoia robotic storage systems, Sparrow robotic arms for item handling, and Robin sortation robots. Amazon has been building this infrastructure since the 2012 acquisition of Kiva Systems for $775 million.

Ocado has built and operates fully automated customer fulfilment centres in the UK, and licenses the Ocado Smart Platform to international grocery retailers. The Ocado grid system handles picking, packing, and dispatch without human pickers in the main operating area.

The IFR World Robotics 2024 report identified logistics as the fastest-growing application segment for mobile robots globally. Deployment has been running ahead of most analyst projections since 2020.

Operators evaluating automation are not making a first-mover bet. They are closing a competitive gap.

Three Robot Categories in Warehouse Picking

Warehouse picking automation resolves into three distinct technology categories with different operating models, capital requirements, and suitability profiles:

Autonomous Mobile Robots (AMRs) — goods-to-person: Robots navigate the warehouse floor to retrieve shelving units or totes and bring them to a stationary human picker. The human does the actual picking; the robot eliminates the walking. This is the dominant current deployment model by unit count.

Picking arms and humanoids — direct item picking: Robots that pick individual items from shelves or totes directly, without human intervention in the pick step. This includes robotic arms (Boston Dynamics Stretch), humanoid platforms (Agility Digit), and vision-guided gantry systems. This category is growing rapidly but is still a smaller deployment base than AMR.

Grid/cube storage systems — fully automated storage and retrieval: Three-dimensional storage grids where robots travel across the top surface, lowering into the grid to retrieve bins. No human access to the storage area. Highest density and throughput per square metre of any system, but requires purpose-built or heavily retrofitted facilities.

AMRs: The Dominant Current Deployment

AMRs are the most widely deployed warehouse automation technology in 2026 by site count, unit count, and operator diversity. They work because they solve the single largest labour cost driver in conventional fulfilment: walking.

In a traditional pick-and-pack warehouse, a human picker walks 15–20 km per shift moving between pick locations. AMR systems eliminate that walking by moving the inventory to the picker. Picker productivity typically increases 2–4x in documented deployments, with variance depending on SKU profile and order structure.

Locus Robotics LocusBot Origin

Locus Origin is a collaborative AMR designed to work alongside human pickers in existing warehouse operations. It does not require warehouse modification — the robot navigates around existing racking, human workers, and equipment using onboard sensors and fleet management software.

The collaborative model matters for operators running existing buildings: retrofitting a conventional warehouse with Locus requires no structural changes and minimal operational disruption. The robot fleet can be scaled up or down, and the system can be deployed in weeks rather than months.

Locus has documented deployments with DHL, Geodis, and multiple 3PL operators in North America and Europe.

Geek+ P800

Geek+ is the dominant AMR provider by unit count in Asian markets and a growing presence in European and North American logistics. The P800 is a shelf-moving AMR — rather than navigating to fixed racking, P800 moves entire mobile shelving units to picking stations.

The pod-based system requires purpose-designed mobile shelving rather than conventional fixed racking, which means conversion from a traditional warehouse involves shelving replacement. For greenfield warehouse builds or full warehouse reconfigurations, this is manageable. For operators converting active operations, the transition requires planning.

Geek+ reports over 50,000 AMR units deployed globally as of 2024.

Compare: Locus Origin vs Geek+ P800

Picking Arms and Humanoids: The Next Wave

The AMR model still requires a human to perform the actual pick. The next category eliminates that step: robots that pick individual items directly. This is harder than it looks. Unstructured item picking — handling arbitrary objects of varying size, weight, orientation, and fragility — is one of the most technically demanding tasks in robotics.

Two commercially deployed approaches exist in 2026:

Boston Dynamics Stretch

Stretch is a purpose-built robotic arm mounted on a mobile base, designed specifically for case handling — moving cases from truck trailers and pallets into warehouse storage. It is not a general-purpose picking robot. It handles mixed-case unloading, a task that is physically demanding, time-constrained, and difficult to automate with traditional fixed-arm systems.

Stretch is deployed with DHL in trailer unloading operations. Its mobile base allows it to operate inside trailer interiors, a geometry that fixed unloading systems cannot address. DHL's deployment represents one of the largest commercial robotics agreements in logistics.

Agility Robotics Digit

Digit is a humanoid robot deployed in warehouse environments by Amazon and GXO Logistics for tote handling — moving totes between conveyor systems, storage locations, and packing stations. Digit does not currently perform individual item picking from unstructured assortments. Its task is structured tote and bin manipulation in defined logistics environments.

Digit is sold through Agility's Robot-as-a-Service model: fleet deployment under a service contract rather than outright purchase. This structures the acquisition as an operating expense and transfers hardware maintenance responsibility to Agility.

Compare: Digit vs Boston Dynamics Stretch

The honest capability boundary: Neither Stretch nor Digit can reliably pick arbitrary consumer items from an unstructured assortment in 2026. Stretch excels at case-level handling. Digit excels at tote-level manipulation. Individual item picking from mixed SKU bins remains a partially solved problem — vision-guided robotic arms with suction and gripper end-effectors handle a subset of items reliably, but not the full range of items in a general retail fulfilment operation.

Grid and Cube Storage: Maximum Density, Minimum Flexibility

AutoStore R5

AutoStore is a three-dimensional storage grid system: a dense aluminium grid structure in which bins are stacked vertically to the full height of the building. Small robots traverse the top surface of the grid, lowering into the stacks to retrieve specific bins and delivering them to port stations where human operators pick the required items.

The R5 is AutoStore's current-generation robot, running on the AutoStore grid with a reported speed of one bin retrieval per robot per approximately 30 seconds. A single AutoStore installation can hold tens of thousands of bins in a footprint that would hold a fraction of that inventory using conventional racking.

Space efficiency is the defining characteristic. AutoStore installations typically achieve 4–5x the storage density of conventional racking in the same building footprint. For operators paying urban or suburban warehouse rent, this translates directly into cost per order.

Hai Robotics HAIO

Hai Robotics' HAIO system uses a different architecture from AutoStore: robots with extending arms that can retrieve bins from specific rack positions rather than digging down through a stack. This offers faster access to any specific bin without the need to move bins above the target, at some cost to storage density versus the AutoStore approach.

Hai Robotics has significant deployments in Asian markets and growing European presence, positioning HAIO against AutoStore in the high-density automated storage segment.

Compare: Hai Robotics HAIO vs AutoStore R5

The constraint that applies to all grid systems: These installations require either purpose-built facilities or significant building modification. The grid structure must be engineered into the building. Operational access for maintenance requires design consideration from the outset. Grid systems are not retrofit options for conventional warehouses in operation — they are capital investment decisions made at the point of a new build or a full facility rebuild.

Which System for Which Operation

The buying decision maps primarily to operation size, order profile, and building situation:

Under 50,000 orders per day, existing building: AMRs are the appropriate entry point. Locus and Geek+ both offer deployments that do not require structural building changes. Capital outlay is substantially lower than grid systems. Deployment timelines are weeks. The productivity improvement is meaningful and the operational risk is manageable.

New build or full facility rebuild, high volume: Model the grid. AutoStore or Hai Robotics grid installations have capital costs that appear prohibitive in isolation, but the operating cost per order at scale — particularly in high-rent markets — frequently justifies the investment over a 10–15 year building life. The operators who made this decision in 2020–2022 are now running cost structures their competitors cannot match with conventional warehouses.

Mixed SKU, irregular items, returns processing: Human pickers alongside robots remain the most reliable solution. AMRs reduce the human labour requirement significantly, but vision-guided robotic picking of arbitrary consumer items is not yet reliable enough for full replacement across all SKU ranges. Deploy AMRs, reduce headcount through attrition and redeployment, and reassess robotic picking capability on a 24-month cycle.

Trailer unloading and case handling: Stretch is the commercially demonstrated solution. If your operation involves high-volume trailer unloading, the case for Stretch is operationally straightforward — it solves a specific, defined, physically demanding task with a commercially proven platform.

Tote-level logistics and structured handling: Digit's humanoid form factor is relevant where the physical environment was designed for human workers — standard conveyor heights, standard tote dimensions, variable spatial layouts. Where wheeled AMRs cannot navigate, humanoids can.

The Cricket's Assessment

> AutoStore is the most capital-efficient warehouse system ever built. It is also the most inflexible — you cannot easily retrofit it into an existing building. The companies that built AutoStore installations in 2020–2022 are now operating at costs that their competitors using traditional warehouses cannot match. The window to make that investment decision is not permanently open. > > The more interesting near-term development is not grid systems — that market is established and relatively slow-moving. It is the trajectory of individual item picking. The gap between what Stretch and Digit can handle today and what a human picker can handle is measurable and is closing. The operators who understand that gap and are tracking it seriously will be better positioned when it closes than operators who are waiting for the technology to mature before paying attention. > > The AMR market is also at an inflection. The technology is mature and the deployment model is understood. The remaining question is consolidation: there are too many AMR vendors for the market size, and several will not survive the next funding cycle. Operator procurement decisions made in 2026 should weight vendor financial stability more heavily than in previous years.

Frequently Asked Questions

How many robots does Amazon use in its warehouses?

Amazon operates more than 750,000 robots across its global fulfilment and logistics network as of 2024. These include Kiva-derived shelf-moving AMRs (Amazon Robotics drives), Sequoia robotic storage and retrieval systems, Sparrow robotic item-handling arms, Robin sortation robots, and Proteus autonomous mobile robots for heavy cart movement. Amazon's robotics deployment is the largest civilian robotics operation in the world.

What is an AMR and how does it work in a warehouse?

An Autonomous Mobile Robot (AMR) in a warehouse context is a robot that navigates the warehouse floor independently using sensors, cameras, and onboard computing. In the goods-to-person model, AMRs retrieve shelving units or tote bins from storage and bring them to a stationary human picker, who selects the required item without walking. AMRs navigate around fixed obstacles, other robots, and human workers in real time. They do not require fixed track or floor modification in most deployments.

What is the difference between AutoStore and a conventional warehouse?

A conventional warehouse uses fixed racking accessed by human pickers or forklift equipment, with wide aisles for access. AutoStore replaces conventional racking with a three-dimensional aluminium grid filled with stacked bins. Small robots travel across the top surface, retrieve specific bins by lifting bins above the target, and deliver them to port stations. AutoStore installations achieve 4–5x the storage density of conventional racking, with no picking aisles required. The trade-off is capital cost and inflexibility — the grid is a fixed installation that cannot be easily reconfigured.

Can robots pick individual items from a warehouse shelf yet?

Robotic item picking is commercially deployed but not universally capable. Vision-guided robotic arms with suction and gripper end-effectors reliably pick a defined subset of items: regular-shaped, non-fragile items with predictable surface geometry. Arbitrary consumer item picking across a full mixed-SKU assortment — including irregular shapes, soft packaging, fragile items, and nested or stacked products — remains partially solved. The capability boundary is advancing, but human pickers alongside robots remain standard for general retail fulfilment operations.

What is Robot-as-a-Service in warehouse logistics?

Robot-as-a-Service (RaaS) is a commercial model in which the robot vendor owns the hardware and the operator pays a monthly or per-task service fee rather than purchasing robots outright. Agility Robotics deploys Digit under a RaaS model. Locus Robotics also offers RaaS deployment. The model converts capital expenditure to operating expenditure, transfers maintenance responsibility to the vendor, and allows fleet scaling without incremental capital commitment. It is increasingly the dominant commercial model for AMR deployment in third-party logistics.

Which warehouse robot is best for a small operation?

For operations below 50,000 orders per day in an existing building, collaborative AMRs such as Locus Origin are the most appropriate entry point. They do not require building modification, deploy in weeks, and deliver measurable productivity improvement with manageable operational risk. Grid systems such as AutoStore require new-build or full-rebuild facilities and substantial capital commitment — they are not appropriate for small to medium operations without those conditions. Direct-picking robots are commercially relevant primarily for operations with specific defined tasks (trailer unloading with Stretch, tote handling with Digit) rather than general picking.

How does warehouse automation affect warehouse jobs?

Warehouse automation reduces the labour requirement per order fulfilled, but does not eliminate warehouse employment at most currently automated sites. AMR deployments typically reduce the number of pickers required per unit of throughput by 30–60%, with remaining staff performing higher-value tasks: exception handling, quality control, replenishment, maintenance. Fully automated grid installations (AutoStore, Ocado) eliminate picking roles within the storage area while maintaining staffing at port stations and in receiving, dispatch, and maintenance. See the Warehouse Picker/Packer profile on the Geppetto Jobs Index for data on displacement risk and workforce transition patterns.

The Geppetto directory tracks warehouse and logistics robots across AMR, picking, and storage categories with full specs and deployment data.