Otto Group and robotics software company Sereact said on October 30 that they will test AI-controlled robots on one of retail’s more awkward warehouse jobs: inspecting and sorting returned customer orders. In a joint announcement dated Hamburg and Stuttgart, the partners said 10 dual-arm systems will be deployed first at an Otto Group logistics center in Poland. In a second phase, they said, 15 more robotic arms may be added based on the initial results, bringing the program to as many as 25. Otto Group said it will evaluate a broader rollout across its logistics network if the second phase is successful.
The companies said returns are hard to automate because items come back folded, crumpled, or repackaged, and because a multi-category retailer’s assortment keeps changing. The robots will run on Cortex, Sereact’s physical AI model, which the company said can recognize and handle items it has not seen before, without product-specific teaching or a fixed item catalog. Cortex, Sereact said, decides in real time how to grasp each item, assess its condition, and route it.
Sereact said Cortex is built on NVIDIA’s physical AI stack. It uses NVIDIA Isaac Sim and Isaac Lab to turn real picks from the returns line into thousands of simulated variants, and NVIDIA Cosmos world foundation models to re-render them across garment states, packaging, materials, and lighting. New policies, it said, are post-trained on NVIDIA Blackwell GPUs and tested against held-out cases before they reach the robots. When a robot struggles with an item, the company said, a tested update can come back within days rather than weeks. Sereact will operate and maintain the systems under a robotics-as-a-service model. Employees, the partners said, will move from repetitive grasping and lifting to supervision, exception handling, and final quality decisions.
“Our customers expect returns to be simple and fast — behind that lies a process that is time-critical and laborious, both for customers and for our business,” said Martin Umland, group vice president of supply chain management at Otto Group. “Returns are notoriously hard to automate, and Physical AI has the potential to change that. This project lets us find out, at real scale, how far Physical AI can take us — an important step towards a more flexible and resilient supply chain.” Dr. Ralf Gulde, chief executive officer and co-founder of Sereact, said, “A robot that only knows yesterday’s catalog is of little use in a returns center. Cortex has to deal with whatever comes out of the next parcel.”
The announcement said the partnership builds on Otto Group’s earlier robotics investments and forms part of the company’s medium-term commitment to invest €350 million in technology and AI. It does not name the Polish site, give a start date or a contract value, or report performance metrics from the trial, which had not yet been described as complete.
What it means for manufacturers
Most midsize manufacturers do not run a fashion returns hall, but many do handle irregular inbound goods: damaged cartons, mixed cases, and items that do not match the packing list. The lesson from a trial like this is less about buying a dual-arm robot and more about whether exception handling, quality checks, and inventory records stay in the same loop when a machine takes the first grasp. That is the same practical question raised in our note on warehouse automation for midsize plants and in recent coverage of DHL’s adaptive robotics pilots.
A practical next step is to walk a day’s worth of messy receipts or returns and write down which steps still require a person’s judgment, and which records have to be updated when an item is diverted. Our recent report on GXO’s Apollo robot trials is another place to compare how operators are staging humanoid and arm-based pilots before any wider rollout.
Questions about exception handling, inbound quality checks, or whether warehouse robotics is worth a closer look?
Contact usSources
Additional sources
- The Otto Group release is the primary account of the trial’s scope, the NVIDIA tools named above, the robotics-as-a-service model, and the €350 million medium-term technology and AI commitment. Independent trade reports published after the announcement largely restated those points and did not add measured results from the Poland site.
Photo courtesy of Otto Group / Sereact, from the companies’ Oct. 30, 2026 announcement.
Try the Statadventure live tools
Inventory tools for US manufacturers: reorder point, safety stock, EOQ and landed cost, with the inputs and assumptions shown.
Open the live tools