Written by Jane Aubrey
Denmark’s Salling Group is trialling humanoid robots from startup Hive Robots in a Føtex supermarket, teaching them to restock shelves mainly overnight, with the aim of autonomous restocking by early 2027.
Most humanoid robot demonstrations happen on a stage, in front of an audience, with a script. The pilot announced in Denmark is deliberately less glamorous: a supermarket, after closing, stacking shelves.
Salling Group, Denmark’s largest retail group, is trialling humanoid robots supplied by startup Hive Robots in one of its Føtex stores. The robots handle restocking, working mainly at night when the store is closed to the public, with staff redeployed toward customer service. The stated goal is for the robots to restock shelves autonomously by early 2027 — meaning that at present they are being trained, not simply deployed. Reporting on the pilot has identified the hardware as Unitree’s G1 platform, a small humanoid that has become a common base for commercial experimentation because of its relatively low price and open interfaces.
The choice of task is smarter than it looks. Night restocking is repetitive, structured and geographically fixed. Shelf positions are known, packaging is standardised, aisles are empty, and the safety risk of a machine working near people is close to zero because there are no people. Compare that with a checkout or a shop floor during opening hours, where a slow or confused robot generates queues, complaints and liability. Retailers have effectively selected the one job in the building where the environment resembles a factory.
That matters because the honest gap in humanoid robotics is not walking, it is reliability under variation. A robot that restocks a known shelf layout at night, at its own pace, with a supervisor on call, is solving a much easier problem than the videos imply — and that is precisely why the pilot is worth watching rather than dismissing. If a humanoid cannot be made economically useful in a closed supermarket at night, the case for deploying it in an open, unpredictable environment weakens considerably.
The economics are the open question. Restocking at night carries shift premiums and contributes to the turnover that plagues retail work, which is the appeal. Against that sit the capital cost of the machines, their throughput relative to a human worker, supervision and maintenance, and the reality that a pilot store with attentive engineers is not the same as a chain of hundreds of stores. Salling frames the programme as easing staff workload rather than replacing staff, and that framing will be tested once the pilot ends.
Europe’s retail labour context makes this sensitive. Automation in Danish supermarkets arrives into a sector with organised labour and comparatively strong protections, which shapes both the pace of adoption and how deployments are described publicly. A pilot announced as workload relief can be experienced differently by the people whose shifts change.
Three things will indicate whether this becomes more than a demonstration. The autonomy rate actually achieved by early 2027, measured without a human intervening. The cost per item handled, compared with a night-shift worker. And whether the approach extends beyond a single store — because the distance between one well-supervised pilot and a fleet across a national chain is where most retail robotics programmes have historically stalled.
