Day: 3 September 2026

The humble cardboard shoebox is often considered to be one of the trickiest problems that warehouse automation technology has yet to resolve, until now, writes Oscar Cutts (pictured, below), Business Development Manager, Nomagic.

Warehouse automation has become remarkably good at moving pallets, cartons and standardized totes. Yet one of the most common items in fashion fulfillment – the shoebox – has remained surprisingly resistant to automation. With shoeboxes making up around 20 per cent of all fashion ecommerce merchandise, even modest improvements in automated handling can have an impact on warehouse productivity.

A shoebox appears to be an ideal object for a robot to handle. It has defined edges, fairly consistent dimensions, and is less deformable than apparel. Unlike sealed cartons though, most shoeboxes consist of two separate pieces: a base and a loose-fitting lid. Slight variations in how the lid overlaps the base, the orientation of the box, or the friction between the two can cause the lid to shift or separate during handling. Warehouses also process hundreds of shoebox designs, sizes, and materials, with inventory continuously changing throughout the day.

One might ask, as many do, why a simple elastic band is not used to prevent the lid separating. This would allow regular suction cup grippers to lift the box from any side and place without risk. Interestingly, following trials and surveys, leading shoe manufacturers and retailers have found elastic bands and other banding methods to be detrimental to the user experience and therefore the brand. So much so, that they dictate specifically that bands must not be used by their distribution partners or anyone else dispatching the boxes.

Shoebox picker is here

This challenge highlights a broader shift across warehouse automation. Instead of relying solely on fixed programming and specialized mechanical tooling, the industry is increasingly turning to Physical AI – robotic systems that combine perception, reasoning, and manipulation to adapt to variability in real time.

A robot must determine what an object is, its position, whether it is stable, where it can be safely grasped, and how that grasp should change based on the object’s characteristics. These decisions must be made in fractions of a second while maintaining human-level throughput.

Modern AI vision systems generate detailed three-dimensional representations of each object, allowing robots to evaluate size, orientation and other physical characteristics before planning a grasp. Machine learning models then determine the most appropriate gripping strategy, while tactile sensing verifies whether the grasp was successful. If conditions change unexpectedly, the robot can reassess the situation and adapt rather than simply fail.

This ability to perceive and respond to real-time variation is becoming increasingly important as retailers seek greater flexibility from their fulfillment operations.

In fashion ecommerce product assortments change constantly, seasonal demand creates swings in inventory, and customers increasingly expect rapid order fulfillment across thousands of SKUs. Automation systems designed for a narrow set of predictable items often struggle in these environments.

One recent example to solve this is Nomagic’s Shoebox Picker, a robotic solution designed specifically to handle two-piece shoeboxes that have historically been difficult to automate reliably. The system combines AI-based perception with specialized end-of-arm tooling that evaluates each shoebox, adjusting its grasp according to the box’s dimensions, lid configuration, and orientation. The robot adapts its manipulation strategy to the specific object in front of it. The system can handle approximately 98% of shoebox SKUs while achieving picking rates of up to 450 shoeboxes per hour.

Shoeboxes may seem like a niche application, but they illustrate a larger trend. As Physical AI continues to mature, robots are becoming capable of handling increasingly diverse products without requiring every object – or every warehouse – to conform to rigid automation rules.

The future of warehouse automation will be driven by smarter robots – systems capable of understanding the physical world well enough to work within its inherent variability.