Robots in hotels: delivery, cleaning and lobby robots, and what to check before buying

A hotel robot is a machine that carries out one narrow physical task inside the building: carrying an item to a room door, cleaning open floor on a fixed route, or answering simple questions in the lobby. Before buying, check the route, the lifts, the Wi-Fi, the staff minutes it still takes, and agree a stop rule for the trial.

A supplier wheels a delivery robot into the lobby, it makes one smooth trip to a suite and back, and the room applauds. The purchase decision is not made in that moment. It is made on a Tuesday morning with forty departures, a service lift out of order and a tray of towels waiting at the station. After reading this article you will be able to define the task in one sentence, check your building against it, and run a trial with a stop rule instead of a demonstration.

What counts as a hotel robot?#

A hotel robot is a machine that performs one repeated physical task inside the building, on a fixed map, with a person nearby to load, unload or rescue it. Three families are sold to hotels today: delivery, floor cleaning and lobby information.

Delivery robots carry an item from a service point to a room door or a station: towels, amenities, a room service tray, a laundry bag. The useful unit of measurement is the trip, not the robot. A machine that completes forty trips a day without help is a different purchase from one that completes forty trips with nine rescues.

Floor cleaning robots scrub or vacuum open floor on a route, usually outside peak traffic. They cover large flat areas consistently and stop at edges, stairs, bathrooms and a room full of luggage. Lobby robots answer simple questions, give directions and sometimes carry a small item or hold a screen. They handle wayfinding and routine questions, not a complaint or a payment.

What all three share: one repeated task, a map of the building, a docking or charging point, and a person who starts and finishes the work. The word AI in the brochure usually means navigation and obstacle avoidance, sometimes speech recognition. Test it like any other AI claim: what exactly does the software decide, what happens when it decides wrongly, and who corrects it.

A self-service kiosk, a smart lock and a chat assistant on the guest's own phone are not robots. Mixing them into one purchase conversation hides the real cost of the physical machine. The decision rule: write the task in one sentence before you look at any product. If the sentence needs the word and, you are buying two things.

What does each type do well, and where does it stop?#

Each type is reliable at one repeated movement and stops at the first thing that needs a hand, a key or a judgement. Knowing where it stops is the specification.

Delivery does well on short, repeated trips between a service point and room doors, especially at peak hours, for items that fit the tray and do not need temperature control. It stops at a closed door, an oversized tray, a guest who does not answer, a lift it cannot call, and a corridor blocked by a housekeeping cart.

Cleaning does well on large open floors with a regular route and consistent coverage. It does not do edges, stairs, wet areas, or a room where luggage and cables sit on the floor. Lobby information does well at wayfinding and the same ten questions. It stops at anything involving money, a complaint, a special request or a language it was not set up for.

The honest measure of any of them is the share of trips that finish without a person touching the machine. Count that in your own building; do not accept a supplier's number for a different building.

Food and beverage delivery adds its own constraints: temperature, allergens, and how alcohol is served and handed over. Those rules belong to your operation, not to the robot. If you serve food and drink, read the same reasoning in our article on AI in hotel restaurants and bars. And remember the arithmetic: a robot that is good at 70 percent of a route still leaves 30 percent with a person, and the remaining part is often the slowest part of the job.

What must the building provide before a robot can work?#

A flat, mapped, connected route and a place to charge. Without those three, the machine stays in the lobby as a demonstration and the staff quietly stop using it.

Start with floors and thresholds: a person walks the intended route with a tape measure and a notebook, and records every threshold, ramp, carpet edge, drain and change of surface. Then doors: width, weight, and whether the door is a fire door that must stay closed or a badge-controlled door the robot cannot open. Each one is a decision, not a detail.

Lifts need either an integration with the lift controller or a written policy on which lift the robot may use and when. Guest lift etiquette and service lift availability both matter. Wi-Fi needs coverage in corridors, service areas, stairwells and the charging spot, not only in the lobby where the demo happened. Charging needs a power point in a place with no guest traffic, enough space to dock, and a named person who plugs it in at the end of the shift.

Safety planning comes next: the fire and evacuation plan must account for a machine parked in a corridor, and staff must know how to move it or switch it off in an emergency. Support is the last item, and it is contractual: a local technician, spare parts and a response time written into the contract, because the machine will stop working at an inconvenient hour.

Who still does the work when a robot is running?#

People do. The robot moves the item; a person loads it, unloads it, cleans it, rescues it when it is stuck and finishes the job at the door.

Loading is a new task at the service point, and it has to fit the existing sequence of a shift, not sit on top of it. Unloading at the room door still needs a person for room entry, for a guest who wants to ask something, and for anything that must be handed over rather than left. For how the shift actually flows, see how a hotel housekeeping department works day to day.

Rescue is the hidden cost: a machine stuck behind a cart or waiting at a lift blocks a trip until somebody walks to it. Count those minutes during the trial. Cleaning, consumables and charging are daily tasks with an owner; if no shift owns them, the machine sits in a corner within weeks.

Peak hours are the test. A robot that works at 11:00 may be useless at 09:00 when the corridors are full of departures and carts. Housekeeping sequencing decides most of this: a delivery robot that interrupts a room sequence costs more minutes than it saves, so the route has to be planned around the floor order the team already uses. The finishing test: for every route, write down what a person still has to do after the robot arrives. If that list is longer than the trip it replaced, the case is weak. Software can help prepare that sequence; our piece on AI in housekeeping separates what software can prepare from what stays with the team.

What will guests ask about cameras, children and accessibility?#

Three questions arrive in the first days: is it filming me, is it safe for my child, and can I get past it with a wheelchair or a pram. Have an answer ready before the machine appears.

On cameras, separate the navigation sensor from any camera that records. Ask the supplier in writing what is captured, whether images are stored, where, for how long and who can see them. A machine that records images in a corridor is handling personal data, and the general duties of protection and notice apply. Describe the duty to your team in plain terms and confirm your own case with a legal adviser before the trial starts.

Notice is simple: a short, plain sign or a line in the guest app explaining what the machine does and what it records is better than a paragraph in the terms nobody reads. On children, think about speed, sensor range, pinch points and behaviour around a moving object. A child will test the machine, so staff need a script for the first week.

On accessibility, a robot parked in a corridor narrows the path for a wheelchair or a pram. The charging spot and the idle parking place must be off the accessible route. Keep the human route open: a guest who does not want a machine at the door should be able to say so, and the team should be able to act on it without a discussion. On language, a lobby robot that speaks one language well is better than one that speaks five badly. Test it with the questions your front desk actually receives.

How do you trial a hotel robot before buying?#

Run one machine on one route for a fixed period, with a baseline measured by hand, a small set of measures and a stop rule agreed in writing before delivery.

Choose one route with real volume and a clear start and end point, not the route that flatters the machine. Measure the baseline by hand for a full week: trips per day, minutes per trip, trips that failed, and staff minutes spent on the task. Use the same measures during the trial, logged by the same person, so the comparison means something.

Include a full weekend and a full housekeeping cycle, and deliberately test the failure modes: blocked corridor, lift out of service, guest who does not answer the door. Hold a five-minute review each day with the shift that owns the machine. Problems surface in conversation long before they appear in a report.

Write the stop rule before the trial: a safety incident, more staff minutes than the baseline, or the share of completed trips falling below the line you agreed. Any one of these ends the trial without an argument. Agree in advance what happens to the machine and the data at the end of the trial, whether you buy or not. The same structure, scope, baseline and stop rule, is set out in our guide to running an AI pilot in a hotel.

What should you ask a robot supplier?#

Ask about the building, the failure, the money and the exit, in that order. The answers tell you more than any demonstration in a lobby.

Building: which lifts, doors and floor surfaces this model has worked with, and whether the supplier will walk your route before quoting. Failure: what happens when the machine stops at two in the morning, who is called, and how long the response takes under the contract. Support: local technician, spare parts, software updates, and who trains your staff, including new staff after the first training.

Money: what is included in the price and what is a recurring subscription, and what the cost is per year after the warranty ends. Data: what the machine records, where it is stored, whether you can export the logs, and what happens to that data if you end the contract. Exit: the terms of the trial, the return arrangement, and the notice period. A machine that is hard to remove is a decision you make once.

Finally, references: ask to visit a building with a corridor width and lift arrangement close to yours, and speak to the person who actually runs the machine, not only the manager who signed. For the broader set of questions that apply to any AI purchase, see our list of questions to ask an AI vendor before buying for your hotel.

What to do next#

Before you accept a demonstration, walk one real route with a stopwatch and a notebook, write down every threshold, door and lift decision, and record what a person still has to do at the end of each trip. That page is your baseline, your specification and your stop rule. The task defines the machine, the building decides whether it can work, the people decide whether it saves anything, the guests decide whether it stays, and the trial decides whether you buy.