Since June, Danish startup Hive Robots has been piloting Unitree humanoid robots inside a Føtex supermarket near Copenhagen, according to Les Numériques. It is a live, named retail deployment, not a staged demo, and the details are a useful check on what humanoid robots can actually do in a store today.
The robots are slower than a human employee and are still guided remotely rather than operating on their own, Les Numériques reports. The pilot's goal is to restock certain shelves unsupervised, a milestone it has not yet reached.
What "pilot" means here
That distinction matters for anyone reading pilot announcements as a signal of market readiness. A remotely guided robot is not an autonomous one: someone is still watching a feed and making decisions, even though the robot carries out the physical task. That can be a reasonable way to de-risk a deployment and collect training data, but it is a different cost and staffing model than a system that restocks shelves overnight with no one in the loop.
Buyers evaluating humanoids for restocking, picking or similar repetitive tasks should ask suppliers directly whether the robot operates autonomously today or is still steered remotely by a person. The answer changes the staffing math: a robot that needs a remote operator still carries a labor cost, just a different one. The humanoid comparison page and price tracker are a reasonable starting point for checking what is actually shipping against what a vendor's roadmap promises.
None of this makes the Føtex pilot a failure. A supermarket running humanoid robots on real shifts, even slowly and under supervision, is a genuine deployment. But it is not yet the unsupervised, full-speed operation that would justify treating humanoids as a drop-in replacement for stocking staff. Tracking its progress toward that unsupervised milestone is the detail worth watching, not the fact that robots are on the shop floor.
The technology that would close the gap
Three stories reported by Robotics & Automation News point at the same bottleneck the Føtex pilot is running into: getting robots to handle objects and learn physical skills without a human closely involved.
Tesla has filed a US patent application for a touch-sensitive "skin" that can be molded around robot fingers and palms to help detect contact and force. Tactile sensing of this kind is a prerequisite for delicate, unsupervised object handling. A patent filing is not a shipped hand, but it shows where Tesla sees the gap.
Skild AI demonstrated a self-play training approach, using its S1 robotics foundation model, that lets robots develop physical skills such as soccer without relying on human demonstrations. It is a research result, not a deployed capability, but it is aimed at the same problem: robots that learn without a human teleoperator or demonstrator in the loop.
Pudu Robotics, known for delivery and cleaning robots in restaurants and hotels, is expanding into semi-humanoid robots and quadrupeds. That is a vendor with an established base in simpler service robots moving into more general-purpose hardware, worth tracking on the best humanoid robots page.
What keeps a fleet running
One more operational detail applies regardless of how autonomous a robot eventually becomes. The Robot Report notes that opportunity charging (topping up batteries during natural pauses rather than taking a unit fully offline) is becoming a key factor in keeping autonomous robot fleets efficient as deployments scale. It is unglamorous, but it is the kind of detail that separates a pilot from a fleet that can run a full shift schedule.
The takeaway for buyers
The Føtex pilot is the most concrete data point available on what a humanoid robot can do in a working supermarket today: real shifts, real shelves, but slower than a human and still remotely guided rather than autonomous. The Tesla, Skild AI and Pudu stories show the component technologies (tactile sensing, self-taught physical skills, broader hardware lineups) advancing. Advancing technology and a deployed, unsupervised robot are not the same thing yet. Anyone budgeting for humanoid robots this year should weight the Denmark result over any lab demo: it answers the question that determines whether a robot is ready to buy, not whether it looks ready on video.