Short answer: an autonomous mobile robot (AMR) automates transport — the walking between tasks. The larger shares of warehouse labour sit in four other layers: storage and retrieval (how densely goods are stored and how fast they reach a person), sorting (dividing flow by destination), piece picking (grasping individual items), and packing and palletising (turning picked items into shippable units). Automating beyond the AMR means choosing which of these layers is your actual bottleneck — and the answer is determined by your order profile, not by vendor brochures.
Layer one: storage that comes to the picker
Goods-to-person systems invert the warehouse: instead of pickers walking to shelves, inventory travels to ergonomic workstations. The main families are shuttle systems running in rack lanes, vertical storage modules, case-handling robots that climb racks, and cube storage — the grid of stacked bins served from above that AutoStore defined as a category. France's Exotec takes a different route with its Skypod robots climbing racking to bring bins to stations. These systems buy two things at once: pickers stop walking (the largest single time saving in manual fulfilment) and storage density rises because aisles shrink. They are infrastructure, not accessories — the decision shapes the building for a decade, which is why this layer deserves the most conservative evaluation of any on this page.
Layer two: sorting, the quiet throughput ceiling
Between picking and packing, flow must be divided: by order, by route, by carrier. Cross-belt and sliding-shoe sorters handle this mechanically at high rates; tilt-tray and bomb-bay sorters serve gentler goods; and a newer family of small mobile sorting robots trades peak rate for modularity — capacity grows by adding units, not by extending a line. Sorting is rarely the first automation bought and frequently the ceiling discovered afterwards: a warehouse that doubles its picking without touching sorting has simply moved its queue.
Layer three: piece picking, the hardest grasp in the building
Picking individual items — especially the mixed, soft, reflective inventory of e-commerce — is the layer where 3D vision, gripper design and AI perception meet their hardest test. The technology works today inside boundaries: defined item families, singulated presentation, tolerant grippers, and a human exception station for the residue the cell cannot handle. Amazon's publicly described systems work — its Sparrow system, announced in 2022, identifies and handles individual products — illustrates the shape of the problem at the largest scale: the value is enormous, and so is the engineering required per item family. For most operators, the sensible entry point is a bounded assortment (health and beauty, packaged goods, apparel in polybags) with success measured item family by item family.
Layer four: packing, labelling and palletising
The end of the line is the least glamorous and often the most reliable automation win: carton erectors and sealers, auto-baggers, print-and-apply labelling, checkweighing, and palletising cells — conventional arms, cobot palletisers for lighter cases, and layer-forming systems at high volume. These processes have stable, well-understood mechanics, which is precisely why they automate predictably. Depalletising incoming goods belongs to the same family and pairs naturally with the vision systems covered in the picking layer.
The layer nobody sells you: orchestration software
Every layer above must be conducted. Warehouse execution software allocates work between robots and people, balances queues so the AMR fleet does not starve the picking stations, and degrades gracefully when something stops. When evaluating vendors layer by layer, ask each one the same question — what happens to your layer when the neighbouring layer stops for ten minutes? — and compare answers. Integration behaviour, not peak rates, is what a peak season actually tests.
Choosing by order profile
| Your dominant problem | Layer to automate next |
|---|---|
| Pickers spend the shift walking | Goods-to-person storage, or AMR transport as the lighter first step |
| Building is full; rent or expansion is the alternative | High-density storage (cube or shuttle systems) |
| Orders wait for carrier sortation cut-offs | Sorting capacity |
| Picking labour cannot be hired at any wage | Robotic piece picking for the bounded item families; keep human exception handling |
| Packing benches are the queue before dispatch | Packing, labelling and palletising cells |
| Everything is "fine" until peak week | Orchestration and exception design — your bottleneck is coordination, not machinery |
Most of the vendors behind these layers are listed in one place: the directory's complete robots and platforms section includes the warehouse-automation specialists alongside the robot makers, and its system integrator section covers the firms that combine layers into a working building.
Frequently asked questions
What warehouse processes can be automated besides transport?
Essentially all of them, at different maturity levels: high-density automated storage and retrieval, goods-to-person presentation, sorting by destination, robotic piece picking of individual items, case and pallet handling, packing, labelling and palletising. Transport by AMR is usually automated first because it is the easiest to add without rebuilding the warehouse — not because it is where most labour sits.
What is goods-to-person automation?
A family of systems — shuttle-based storage, cube storage, vertical lifts and case-handling robots — that bring inventory to a stationary picker instead of sending the picker walking to the inventory. It removes travel time, which is the largest single slice of manual picking labour, and it is the layer where storage density and throughput are won or lost.
When does robotic piece picking make sense?
When the item range is bounded enough for grippers and vision to handle reliably, volumes justify a cell that works at a steady rate for long hours, and the alternative is a chronic hiring problem at the pick face. Mixed, fragile or constantly changing assortments remain the hard cases; the honest evaluation is a trial with your worst items, not your easiest.
Do I need to automate everything at once?
No — and most successful warehouses do not. The common sequence is transport first (fast to install, easy to scale), then goods-to-person storage where density or walking time bites, then sorting and picking cells at the proven bottlenecks. Each layer should pay on its own numbers, because a warehouse's real bottleneck moves as each layer is automated.