News · 2026-10-12

HIVE’s supermarket humanoid pilot still has a human at the controls

HIVE Robots and Danish retailer føtex are testing humanoid shelf replenishment in a live supermarket with a remote human pilot controlling each robot. HIVE says the operation records movements and store interactions to support learning. The pilot is a real deployment and data-collection effort, but the available primary evidence does not establish autonomous stocking or labor savings.

Key facts

• HIVE founder Jakob Sand describes the current staffing ratio as one pilot for one robot.
• The partnership was announced March 11; an October 11 report brought renewed attention to the ongoing pilot.
• The companies are starting with shelf replenishment after stores close.
• The primary sources are HIVE’s retail page and its føtex partnership announcement.

The store is an important part of the experiment. A robot has to handle products, shelves and access routes built for people, rather than a specially arranged industrial workcell. HIVE and føtex are testing whether a human-shaped machine can fit that environment. The retailer’s rationale is plausible, but the pilot has not yet supplied a public comparison proving humanoids outperform other designs.

HIVE’s company post describes the present operation most directly. A person wears a virtual-reality headset and uses two controllers to operate a full-size robot in the store. The system records movements, which items are placed and which shelves are reached. HIVE says the robots are learning to do more of the work themselves.

This is best understood as demonstration collection attached to a real service trial. A learner driver can operate a vehicle with close instruction before independently handling unfamiliar traffic. The instructor’s presence does not make the vehicle imaginary, but it changes what the observed drive proves. Likewise, remote human control demonstrates that work can be performed through the machine; it does not show the machine independently selected and executed the whole task.

The existing lesson on imitation learning explains how demonstrations can become training data. The dossier does not establish HIVE’s precise training architecture, the volume of recorded data or a measured improvement curve. Those details should not be filled in from a generic account of robot learning.

The partnership’s starting task is overnight replenishment while the store is closed. HIVE’s current retail page says live test sessions are underway. That supports a concrete deployment description. It leaves important operating questions unanswered, including the total number of store units, active hours per night, cumulative pilot hours and measured stocking success.

HIVE labels a future service program “Launch Partner.” Its proposed 2027 phases include one robot with one teleoperator working eight hours nightly, followed by a phase with three robots, two teleoperators and 24 hours of combined nightly coverage. These are plans on a service page, not logged results from the present føtex pilot. Multiplying planned hours into an operating record would materially overstate the evidence.

The distinction also changes the economic story. At one operator per robot, the primary sources do not show a labor-productivity advantage. HIVE’s stated direction is for a person eventually to supervise more robots as automation increases. Whether that improves cost depends on reliability, intervention frequency, supervision requirements and the tasks performed. The checked sources publish no measured cost-per-shelf comparison or third-party safety evaluation.

An October 11 Euronews report, read in the dossier, attributes the machines to Unitree and reports an ambition for independent stocking by early 2027. HIVE’s checked pages do not establish that specific deadline. The viral description identifies a G1 model, but neither the primary pages nor the outlet account verified that model designation. An aspiration, a manufacturer and a model are separate claims.

Stakeholder reception comes chiefly from the pilot’s participants. Sand describes the data and supervision strategy; føtex operations director Simon Kudsk describes the appeal of fitting human-designed spaces. These are informed but interested accounts. The supplied Reddit export confirms that the pilot circulated on October 12, but contains no comments or reliable engagement totals from which to infer public acceptance.

The strongest counterargument to the excitement is that teleoperation can make a robot demonstration appear more autonomous than it is. The primary sources allow that ambiguity to be removed: a human currently supplies the control. The pilot’s real significance is therefore the attempt to turn repeated store work into an eventual scalable operating model. Its next persuasive milestone would be measured autonomous task success and reduced human intervention, with failures and supervision time reported alongside successful shelf placements.

Primary source, verified: read the paper →

Ground Truth. (2026, October 12). HIVE’s supermarket humanoid pilot still has a human at the controls. Ground Truth. https://groundtruth.day/news/hive-foetex-teleoperated-shelf-stocking-pilot.html

@misc{groundtruth:hive-foetex-teleoperated-shelf-stocking-pilot,
  title  = {HIVE’s supermarket humanoid pilot still has a human at the controls},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/hive-foetex-teleoperated-shelf-stocking-pilot.html}
}

Topics: robotics · humanoids · teleoperation · retail · deployment