Industries / Industrial equipment and services / Industrial machinery / Conveyors and material handling
Automated storage
Automated storage is the machinery that puts goods away and brings them back without a person walking to a shelf: cranes in rack aisles, shuttles, carousels, lift towers and robots that run on grids and racks.
5
Milestones achieved, from 1957 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1957
Shelves that come round
A vertical carousel is a tall cabinet whose shelves circle like a Ferris wheel and stop with the wanted one at an opening at working height. A history published by SSI Schäfer, a storage maker, says White Systems introduced its Power File vertical carousel for offices in 1957 and Sperry Rand launched its Lektriever later that year, and that several companies sold vertical carousels by the late 1960s. Hänel, a German maker, says it began series production of filing carousels in Europe in 1957. SSI Schäfer dates the vertical lift module, a tower that stores trays and brings one to an opening on request, to the late 1980s.
Credit
White Systems; Sperry Rand; Hänel
1984
Cycle times by formula
Yavuz Bozer and John White published travel-time models for automated storage and retrieval systems in IIE Transactions in December 1984. The models give the expected travel time of a storage and retrieval machine, which moves along and up the aisle at the same time, for one store or fetch per trip and for a store and a fetch in one trip, so a planner can estimate a system's throughput, and check a maker's figure, without building a simulation. In 2020 Andreas Rücker, Jona Rief and Johannes Fottner of the Technical University of Munich still started their study of stacker cranes from that paper, and compared cranes on the reference cycle of the European guideline FEM 9.851 and the test runs of the German guideline VDI 3561, which let buyers judge machines from different makers on the same moves.
Credit
Yavuz A. Bozer and John A. White
2002
A shuttle on every level
A shuttle system puts a small vehicle on each level of a rack aisle and a lift at the aisle's end, so many bins move at once instead of one crane serving the whole aisle. Knapp, an Austrian maker, says its first OSR Shuttle started work at Richter Pharma in Wels, Austria, in 2002. By 2018 there was enough research on shuttle systems' throughput, energy use and control for Nenad Kosanić, Goran Milojević and Nenad Zrnić of the University of Belgrade to publish a survey of it.
Credit
Knapp; Richter Pharma
2017
Robots in three dimensions
Exotec, a French company founded in 2015 by Romain Moulin and Renaud Heitz, built Skypod robots that move in three dimensions, across the floor and up the racks, to fetch bins, so the system needs no crane fixed in each aisle. Its first permanent installation went into a warehouse of Cdiscount, a French online retailer, in December 2017, VentureBeat reported. Cdiscount's logistics arm, C-Logistics, opened a second system at Réau, near Paris, in 2019 with 48,000 bins and added 13,000 more in 2023; its site director, Stéphane Limpalaer, says extra robots can be added for sales peaks. Exotec said in October 2025 that it had deployed more than 10,000 Skypod robots, by its own count.
Credit
Exotec (Romain Moulin, Renaud Heitz); Cdiscount
2025
A storage maker fails
Attabotics of Calgary, Canada, founded in 2016, built a dense storage structure in which robotic shuttles moved both across and up and down to reach bins, with no aisles. At the start of July 2025 it filed a notice of intention for bankruptcy protection and laid off most of its roughly 200 staff; its filings listed $32 million in assets, including 160 patents, against $73.5 million in liabilities, after more than $194 million (US) raised, The Robot Report wrote. LaFayette Systems, a material handling group from Kentucky, bought the company in September 2025 and restarted it in February 2026.
LessonA storage structure that only its maker can extend and service ties every customer to that maker's finances.
Credit
Attabotics; The Robot Report (reporting); LaFayette Systems
Watch for
One port for every bin
AutoStore says FlexBinsPort, due in the fourth quarter of 2026, lets one workstation handle every standard AutoStore bin height, and estimates that it raises storage use by 10 to 15%. It also estimates that CubeFleet, its new maintenance alert system, could cut unplanned downtime by 35 to 45%.
Achieved whenA customer publishes storage use and downtime at a named site, measured before and after the change.
Claimed by
AutoStore, October 2026
Open
01
Energy per bin, robots included
A published measurement of the electricity that a robot grid system and a rack-going robot system use for each bin stored and fetched, taken with the method already used on older machines, so that buyers can compare every type in one table.
The briefCredit · unclaimed02
Throughput claims, checked in public
An open-source simulator of grid storage, in which robots on top of the grid lift off the bins above the one they need (called digging), that takes a site's grid size, bin count, robots, ports and order mix and predicts bins handled per hour; then a check of its predictions against figures published for real sites.
The briefCredit · unclaimed
Open 01 · the brief
Energy per bin, robots included
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A published measurement of the electricity that a robot grid system and a rack-going robot system use for each bin stored and fetched, taken with the method already used on older machines, so that buyers can compare every type in one table.
Why it is still open
Thomas Stöhr, Michael Schadler and Norbert Hafner of Graz University of Technology measured three types on the market, a mini-load crane, a horizontal carousel and a shuttle system, with one test cycle and load pattern, and published the results in 2018. The work on robot grid systems that a search of OpenAlex found on 19 November 2026 estimates energy by simulation, as in a 2022 study by Xingwei Chen, Peng Yang and Zixin Shao, not by measurement at a working site.
Built with what exists
The Graz benchmarking procedure (a set load pattern of 90% and 50% payloads and standby time, and a reference double cycle); clamp-on power loggers fitted by the site's own electrician, or the system's built-in power meters; Python for the analysis.
First step
Ask Graz University of Technology's Institute of Logistics Engineering for the full test protocol, and ask two operators of robot storage systems for a four-hour test window.
Done looks like
Watt-hours per double cycle (one store and one fetch) for at least two robot systems at working sites, under the Graz load pattern, published with the raw logs beside the 2018 figures for cranes, carousels and shuttles.
Open 02 · the brief
Throughput claims, checked in public
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source simulator of grid storage, in which robots on top of the grid lift off the bins above the one they need (called digging), that takes a site's grid size, bin count, robots, ports and order mix and predicts bins handled per hour; then a check of its predictions against figures published for real sites.
Why it is still open
The throughput figures in case studies come from makers and their customers. Researchers model these systems, for example Anni Yue and Stephen Smith in their 2023 study of digging, run on "a custom-built discrete event model", but no model checked against named installations turned up. A GitHub search on 19 November 2026 for "autostore simulation" found four small student repositories, none with a star or a comparison with a real site.
Built with what exists
Python with SimPy; the digging rules described by Yue and Smith; published case studies that state grid size, robot count and throughput; a coding agent to write and test the model under a person's review.
First step
Collect 20 published case studies that state grid size, robot count, ports and bins per hour, and put them in one table.
Done looks like
The model predicts the published throughput of five named installations within 15%, with inputs, code and errors published under an open license.