Humanoid robots in retail: AIn-Guard’s Physical AI concept

AI-generated concept illustration.

In a small shop, a customer looking for help may have to wait for the person serving at the checkout. For the owner, that is a familiar limit: there is only so much attention to go around. A robot that could offer useful assistance would be valuable here. Making that interaction work in an ordinary shop is a substantial research challenge.

At AIn-Guard, we are considering how humanoid robots could work alongside camera-based video analysis. The aim is to explore support for customers and an additional source of information for staff. This is a research concept; AIn-Guard does not currently offer a humanoid robot service.

A useful conversation takes more than a correct answer

Giving someone directions sounds like a modest task. A robot would first have to recognise that the person is speaking to it, understand the request over shop noise and judge when to reply. A product name may be unclear. The customer may interrupt, change their mind or need a member of staff instead.

These are research problems in human–robot interaction. For example, a 2025 field experiment in a shopping mall investigated how a dialogue robot could manage conversational timing amid background noise. Even this narrow part of giving directions required its own system design and evaluation.

For a retailer, the benefit would be a customer getting help at a moment when staff are occupied. That benefit would have to be demonstrated: did the person find what they needed, was the exchange comfortable, and did it leave the team with less work or more? Those questions should guide the design from the start.

Where fixed cameras and a mobile robot might complement each other

Fixed cameras observe a scene over time from a consistent position. A mobile robot could offer a view from somewhere else. Combining those capabilities is one of the ideas behind AIn-Guard’s Physical AI concept.

The timing matters. A robot arriving after an event would show the scene as it is then. It cannot recover an earlier action that no camera recorded. Its possible value would be in observing an ongoing situation or checking a defined area where a new view is still useful.

That leaves a practical question to investigate: would moving a robot there add useful information quickly enough, compared with another fixed camera or a member of staff? A crowded aisle, an obstructed route or an event that has already ended could make the extra view unhelpful. Staff would remain responsible for assessing any footage and deciding how to respond.

A humanoid has to earn its place in the shop

We are interested in the humanoid form because it offers a way to combine movement, gestures and conversation in the same machine. Its usefulness would depend on the task and the people around it. The MuMMER research project, which deployed a social robot in a shopping mall, brought together perception, conversation, spatial reasoning and motion planning. Customer assistance involves the interaction of all these capabilities.

In a small store, the available space matters too. A robot would need to share narrow routes with customers and staff, cope with changes in the layout and behave predictably when it cannot continue. Maintenance, charging and the time spent supervising it would all affect whether it belongs in that shop.

A worthwhile result would be something the owner and customers can notice: an assistance request handled well or a useful view obtained without disrupting the store. We would need evidence of that benefit before making a case for routine deployment.

The starting point is the retailer’s work

AIn-Guard’s current camera-based approach analyses images from connected cameras for unusual movement patterns and sends short clips to authorised staff. Our article on the technology behind AIn-Guard explains how movement analysis leads to a video alert. A future robot would bring additional questions about interaction, movement and the data it collects. Those would need to be assessed in their own right.

For the Physical AI concept, the next useful conversations are with retailers about specific tasks: where customers wait for help, where staff lack a useful view and what assistance would actually improve the working day. A narrowly defined task would give researchers and the store team something concrete to develop and evaluate together.

If you run a shop and see a task worth exploring, discuss it with AIn-Guard. Mention “Physical AI” in your enquiry and describe the situation in your store.