Museum Robots Split into Exhibit Buys and Operations Buys

The Two Museum Robot Briefs

A museum buying a robot is doing one of two things, and the paperwork differs for each. An exhibit robot is a display purchase, funded from the exhibition budget, approved by curatorial staff and measured in visitor numbers. An operations robot is a facilities purchase, funded from running costs, approved by the head of estates and measured in floor area cleaned.

The two briefs pull the hardware in opposite directions. An exhibit robot has to hold attention while standing still, so the money goes into an expressive face, a voice and a conversation system. An operations robot has to run for hours with nobody watching, so the money goes into battery capacity, a charging dock and reliable floor mapping. A machine built for one job is usually mediocre at the other.

Read every vendor demo against the correct column before anything else. A humanoid filmed answering visitor questions in a quiet studio is being sold as an exhibit, whatever the slide deck says about staff efficiency. The same split runs through the rest of the service robots hub.

Guide and Docent Robots on the Gallery Floor

A guide robot leads visitors between exhibits, stops at each one, and delivers a short piece of interpretation. The longest-running public example is Lindsey, built by the Lincoln Centre for Autonomous Systems at the University of Lincoln and deployed at The Collection in Lincoln. Over one seven-month stretch the project reports that Lindsey travelled close to 300 km and delivered more than 2,300 guided tours.

Convert that into tours per day. Roughly 2,300 tours across seven months is about eleven a day, which is more than most volunteer docent rotas cover. It also shows the machine ran unattended long enough to count. A robot that needs a technician standing beside it is a demonstration rather than a docent.

The Smithsonian ran the best-known short pilot. In 2018 it placed Pepper units, donated by SoftBank Robotics, in six of its own spaces. One site used them to translate Kiswahili phrases in an African art gallery. SoftBank Robotics ended new Pepper production in 2021, so anyone copying that pilot now is buying used hardware or moving to another platform.

Telepresence Robots for Visitors Who Cannot Attend

A telepresence robot is a screen and camera on a driveable base that a remote visitor steers through the galleries. The visitor sees the room, hears the guide, and appears on the screen so staff can talk back to a face. Hastings Contemporary in the UK ran the most widely reported programme of this kind, offering robot tours to people who could not travel.

This is the one museum robot category serving people who cannot be reached another way. A recorded video tour is a broadcast that answers nothing. A telepresence robot lets someone stop at the painting they chose, ask about it, and get a reply from a curator. Schools use the same setup to reach collections they cannot afford to bus children to.

The constraints are dull and they decide the project. The robot needs reliable Wi-Fi in every gallery and a staff member to launch and dock it. It also needs a booking calendar, because only one remote visitor can drive it at a time. That single-driver limit caps the audience, so fund telepresence as an access programme rather than as a way to lift visitor numbers.

Cleaning and Patrol Robots after Closing Time

Floor scrubbing is the museum robot job with arithmetic a finance officer can check. Avidbots and Tennant both sell autonomous scrubbers that map a building once and then clean the same route nightly. A large museum atrium is close to the airport concourse those machines were built for, and the specifications are set out in the cleaning robots guide. What you save is a cleaner's overnight shift, so compare the lease against that wage first.

Museum floors carry restrictions an office floor does not. Conservation staff limit which chemicals touch stone and timber, hold humidity inside a set range, and often bar wheeled equipment from named galleries. Any scrubber needs no-go zones mapped around plinths and cases. A machine that clips a plinth once is barred from that room for good, so agree the exclusion map with the conservation team before the first survey.

Overnight patrol is the other after-hours job. Knightscope sells autonomous patrol machines that record video and flag people in spaces that should be empty, which suits a large site with long corridors and two night staff. It does not replace the intruder alarm and access control a museum insurer already requires. Price it as an addition to that system, and check what your policy actually demands using the figures in the security robot cost breakdown.

The Robot That Is Itself the Exhibit

Some museums buy a robot because the robot is what visitors came to see. Ameca, built by Engineered Arts in Cornwall, is the machine most often bought for that job. It holds a conversation, tracks faces and pulls expressions. Units are installed at Dubai's Museum of the Future, the Computer History Museum in California and the National Robotarium in Edinburgh.

The buying questions here have nothing to do with productivity. An exhibit robot that stands dark for a fortnight costs the visitor experience team a booked school group, which a parked floor scrubber never does.

Content upkeep is the recurring cost exhibit buyers miss most often. A conversational robot running on a large language model needs its answers curated and re-checked. Otherwise it improvises museum facts that are wrong, in front of the people who trust you for them. Budget staff time each season, the same way you budget for rewriting label copy. Put four questions to any exhibit-robot supplier before signing.

• How many hours a day the machine can run with visitors in front of it
• What the spares and service agreement covers, and the response time
• Who writes and refreshes what the robot says
• What the gallery looks like on a day the machine is switched off

Where Museum Robots Fail

Speech recognition is the first thing a gallery breaks. Far-field microphone arrays separate one voice from background noise, and a hard-surfaced gallery holding a school group is among the noisiest rooms a robot will meet. Researchers writing in Applied Sciences in 2026 tested how a docent robot's voice, its appearance and the gallery's ambient sound changed how much visitors liked it and how satisfied they were with the interaction. If your galleries are loud, specify a touchscreen fallback so a visitor can tap the question the robot missed.

Navigation is the second. A robot that maps a quiet gallery at 8am meets a different room at 2pm on a wet Saturday. The floor is then covered in pushchairs, bags and people standing still to read. Guide robots slow down and ask people to move, which works until the crowd is dense enough that the machine stops and waits.

The third failure has no engineering fix. A warehouse robot that stalls is watched by three people who are paid to be there. A guide robot that stalls mid-tour is watched by forty visitors, several of whom will film it. What we see readers get wrong most often is filing that under public relations. It belongs in the specification. A museum should hold a guide robot to a higher reliability bar than a distribution centre holds its pickers.

Who Should Skip Museum Robots Entirely

A museum running on a handful of staff and a volunteer rota should buy none of these. Every category above assumes somebody charges the machine, checks it each morning and rings support when it stops. Nobody has that hour. Spend the money on interpretation panels, seating and a better audio guide instead.

Skip the guide robot as well if your galleries have steps or thresholds a wheeled base cannot cross. Mapping software does not solve a stair. Telepresence and cleaning survive an awkward building better, because both can be confined to the floors that work.

Two changes would move this advice. A guide robot that held a natural conversation at gallery noise levels would turn the category into a staffing tool, and nothing shipping today does that dependably. A fall in exhibit-grade humanoid prices to roughly what a large interactive display costs would open the exhibit case to regional museums.

• Volunteer-run sites with nobody assigned to daily robot upkeep
• Galleries with steps or narrow thresholds a wheeled base cannot cross
• Museums expecting a robot to replace front-of-house staff rather than absorb repeated questions
• Anyone funding an exhibit robot without a budget line for refreshing what it says