IN Brief:
- CEVA has deployed Sereact robots with two arms for Zalando returns at Greven in Germany and Świebodzin in Poland.
- Sereact’s Cortex platform identifies, grasps and sorts previously unseen returned products without requiring training for each individual article.
- The live systems are delivered through a robotics as a service model and form the first operational stage of a wider European returns automation programme.
CEVA Logistics has put Sereact robotic systems into live operation for Zalando returns at logistics facilities in Greven, Germany, and Świebodzin, Poland, targeting a process where product condition and presentation vary too much for conventional fixed programming. The robots use two arms and Sereact’s Cortex platform to grasp, identify and sort returned fashion items without requiring a separate training routine for every article.
Returned garments present a more difficult automation problem than new products picked from structured warehouse storage because they can arrive unfolded, loosely packed, partly wrapped or positioned very differently from their original outbound presentation. Fabric also changes shape when compressed or lifted, leaving the vision system without the predictable edges and dimensions available when a robot handles a rigid carton.
Cortex responds to that variation by interpreting the object presented to the station and selecting how it should be grasped and sorted, while the two arm arrangement allows the robot to reposition a garment after the first pick. When fabric is folded over itself or an identifying feature is obscured, the item can be manipulated before the next decision rather than immediately being diverted to an operator.
Human intervention remains necessary for condition assessment, resale decisions and unusual damage, so CEVA is shifting the division of work rather than attempting to automate every return. Employees can supervise stations, manage exceptions and perform quality control while the robots take over a larger share of repetitive lifting and sorting.
Supplying the system through a robotics as a service model changes the commercial structure alongside the physical process. CEVA can use the equipment and software as an operating service instead of purchasing the complete automation system outright, while Sereact retains an ongoing role in software updates, maintenance and system performance as deployments expand.
That arrangement still leaves the robotic station dependent on the rest of the returns line. A completed pick creates little value when inbound containers arrive irregularly, inspection work is delayed or downstream conveyors cannot accept sorted items, because the station will eventually wait for another process regardless of its own cycle speed.
Exception handling creates the same dependency at another point in the workflow. If unusual presentation, damage or identification failures send too many garments to manual processing, employees can become the new constraint even while the robot handles straightforward products efficiently. Useful capacity therefore depends on the proportion of the total returns stream that can move through the automated path without intervention.
CEVA already handles returns within a wider Zalando logistics relationship that runs alongside fulfilment and inbound activity across several European markets. The companies extended those logistics agreements through 2030 in July, giving the current automation programme an established network in which additional installations could be introduced if the first two sites perform as intended.
A returned garment moves through the warehouse in the opposite direction from a conventional outbound order. Instead of a known item leaving controlled inventory for a customer, the site receives goods from many customers in varying condition and has to establish identity, quality and destination before the product can return to sale or move towards another disposition.
Delays during that sequence reduce the amount of time commercially usable stock remains available for resale. Fashion inventory can lose value as seasons, promotions and customer preferences change, so shortening the interval between physical receipt and inventory reintegration can increase the chance of selling a returned item at its intended price.
The Greven and Świebodzin installations target part of that interval by automating identification, manipulation and sorting after the product reaches the robotic station. Their contribution to the wider process will be determined by how frequently Cortex can resolve an item without intervention and whether inspection, conveyors and downstream handling can absorb the resulting flow.
Operational data from those stations can also help identify where the process breaks down. A recurring exception associated with a particular type of garment, packaging method or presentation gives the operator evidence that the problem may sit upstream of the robot, allowing changes to inbound handling or workstation design rather than treating each failed pick as an isolated event.
Consistent robotic handling can provide similar visibility into throughput. Station data can show completed picks, exception volumes and periods of inactivity, helping operators distinguish between limitations in the robot itself and delays created by another part of the returns line.
Zalando had already strengthened its relationship with Sereact before the deployments entered service by joining the company’s Series B financing as a strategic investor in July 2026. Sereact says more than 200 Cortex systems operate across Europe and have completed more than one billion production picks, giving the underlying platform a wider installed base even though fashion returns introduce different handling challenges from conventional warehouse picking.
Operating the technology at two sites now gives CEVA, Zalando and Sereact comparative data across different warehouse environments. Product mix, labour patterns, conveyor layout and local process design can all influence performance, so extending the system more widely will depend on whether the underlying approach remains effective as those conditions change.
The commercial test is therefore the performance of the complete returns process rather than the novelty of the robot. A wider rollout becomes credible when identification and manipulation remain reliable across changing products, manual exceptions stay manageable and the surrounding warehouse can convert faster sorting into faster reintegration of saleable stock.
