Returns Processing Robot: ABB AI Picking Meets Returns

Every online order has a chance of coming back, and the box that comes back is harder to handle than the one that went out. A returns processing robot has to deal with items it has never seen, in packaging nobody planned for, and decide where each one goes next. This post looks at how returns were handled in the mail order era, how reverse logistics became its own business, and what AI picking robots, including ABB's Item Picking modules, now do in returns centres. It ends with the problems that are still open.

Conveyors and the packaging and processing area at the Amazon fulfillment center in West Kaiserslautern, Germany, May 2025. Photo: U.S. Army 405th AFSB / Cameron Porter via Wikimedia Commons (public domain)

From catalogue houses to the first returns centres

Returns are as old as buying things you have not seen. Aaron Montgomery Ward started his mail order business in Chicago in 1872, and by 1883 his catalogue ran to 240 pages and 10,000 items. Richard Sears put out his first general catalogue in 1896. In 1908 Montgomery Ward opened a 1.25 million square foot Catalog House along the Chicago River. Clerks filled orders in buildings like this by hand, and they handled whatever customers sent back the same way.

The former Montgomery Ward Catalog House on the Chicago River, opened in 1908, photographed in 2007. Photo: TonyTheTiger / Wikimedia Commons (CC BY-SA 3.0)

Shipping a parcel back got easier when the US Post Office opened its Parcel Post service in 1913, so a customer could post a package back as easily as the store had posted it out. Catalogue firms grew quickly once rural customers could order and send back goods through their local post office. For most of the century, though, returns stayed a side job. A clerk at a returns desk unpacked each item, judged its condition by eye, filled in a form and pushed it to a shelf, a repair bench or a bin for scrap.

Postmaster General Frank Hitchcock at the opening of the US Parcel Post service, 1913. Photo: National Photo Company Collection, Library of Congress via Wikimedia Commons (public domain)

The work got a name in 1992, when James R. Stock wrote a white paper called Reverse Logistics for the Council of Logistics Management. Barcode scanning and warehouse software then made it possible to track a returned item back to its order. Conveyors and automatic sorters, first built for mail and outbound parcels, were moved into dedicated returns centres. What stayed manual was the hard part: opening a bag, looking at the item and deciding if it could be sold again. Every returned item still passed through a person's hands at that point.

Clerks sorting mail by hand at the General Post Office in Wellington, New Zealand, in the 1900s. Photo: Archives New Zealand / Wikimedia Commons (CC BY-SA 2.0)

AI picking robots enter the returns centre

Returns are now a large and growing problem. Modern Materials Handling asked in September 2026 whether the industry can "finally figure out returns". On 30 September Robotics & Automation News ran a piece by Pramod Ghadge of Unbox Robotics. He cites a survey of 491 executives by Bain, the World Economic Forum and the University of Cambridge in which 79 percent called circularity important but only 20 percent felt their supply chains were ready. His point is that returned volume is erratic, and every item needs its own decision: restock, refurbish or dispose.

Overhead conveyors running through an Amazon fulfillment center in West Kaiserslautern, 2025. Photo: U.S. Army 405th AFSB / Cameron Porter via Wikimedia Commons (public domain)

ABB has put returns into its own pitch. In March 2025 it added two modules to its Item Picking family. The Robotic Fashion Inductor picks polybagged clothes and accessories and places them on a sorter at up to 1,300 picks an hour, and the Robotic Parcel Inductor handles boxes, bags and envelopes at up to 1,500. ABB quotes picking accuracy above 99.5 percent and says handling random items helps with returned packages, which arrive with no fixed time, type or date. The vision side is explained in our post on 3D vision picking.

ABB's Robotic Fashion Inductor picking a polybagged item for a sorter. Photo: ABB

The original ABB Robotic Item Picker, launched in 2023, uses AI vision and a suction gripper to pick unknown items from mixed bins at up to 1,400 items an hour. In a returns centre that is the step after inspection, when an item has been accepted and has to go back into a storage tote. Coordination is moving to software too. Destro AI raised $8 million in late September to run mixed robot fleets from one system, and is piloting it with Yusen Logistics. Our post on parcel sorting and induction covers the sorter side.

ABB's Robotic Item Picker, a small IRB arm with a suction tool inside a guarded picking cell. Photo: ABB

Where robotic returns go next

For shoppers, the next few years should mean faster refunds and fewer trips to the post office. When a robot can identify and grade an item within seconds of it arriving, the refund can be triggered at once instead of days later. More drop off points will take items with no box, and the sorting moves to a regional centre. Retailers will also use the grading data to spot products that come back too often, such as clothing with a sizing problem.

Visitors watch the processing line at the Amazon fulfillment center in West Kaiserslautern, 2025. Photo: U.S. Army 405th AFSB / Cameron Porter via Wikimedia Commons (public domain)

In warehouses, returns cells will look more like the forward picking lines. An ABB Parcel Inductor or Fashion Inductor can sit at the start, with a Robotic Item Picker putting accepted items back into totes, and Flexley Mover AMRs carrying those totes to storage. Unbox Robotics argues for vertical systems that use the height of a building, so peak season capacity does not need a bigger floor. Lines like these are usually laid out and tested in RobotStudio before any hardware arrives.

ABB's Robotic Parcel Inductor lifting a box with its suction gripper. Photo: ABB

The open problems are mostly about judgment. A robot can pick a polybag reliably, but deciding whether a jacket has been worn or a phone has been tampered with is harder, and a wrong grade costs money in both directions. Cells are still expensive for smaller retailers, and returns volume swings so much that a cell sized for January sits idle in May. Sites also need technicians who can retrain the vision models when the product range changes, and ABB's promise of commissioning in about a week only covers the start. A cobot such as ABB's GoFa CRB 15000 could work at an inspection bench beside people, but the retailer still has to choose which steps are safe to hand over.

The vision screen of ABB's Robotic Item Picker, showing items it has found in two bins. Photo: ABB

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com