Bed Making Robot: From Hotel Maids to ABB YuMi Hands

A bed making robot is one of the hardest home robots to build, because a sheet has no fixed shape and a mattress is heavy. This post looks at how beds were made by nurses and hotel housekeepers, how university labs first taught robots to pull a blanket and fold cloth, and what the September 2026 humanoid demos actually showed. It also covers where ABB's dual-arm YuMi fits, since researchers used it for some of the best known cloth folding work, and what still stands in the way of a robot that strips and remakes your bed every morning.

A freshly made king bed in a hotel room, 2017. Photo: Kurt Kaiser / Wikimedia Commons (CC0)

How nurses and hotel housekeepers made beds by hand

Bed making grew up in hospitals, barracks and hotels, where it had to be fast and look the same every time. Nurses and military recruits learned hospital corners, a tight fold that keeps the sheet tucked under the mattress. Speed became a point of pride. Guinness World Records lists 14.0 seconds as the fastest time for two people to make a bed with blanket, sheets, pillow, counterpane and hospital corners, set in 1993 by two nurses from the Royal Masonic Hospital in London. Even the sheets changed to save time. In the 1950s Bertha Berman filed a patent for a fitted sheet with elastic corners that holds itself on the mattress.

Vassar College students learn to make hospital corners during wartime nurse training, 1917 to 1918. Photo: US National Archives via Wikimedia Commons (public domain)

In hotels the job fell to room attendants, and it was hard on the body. Lifting mattress corners to tuck sheets, bending over beds and pushing heavy linen carts add up across a shift. A 2010 study in the American Journal of Industrial Medicine looked at OSHA injury logs from five unionized US hotel companies over three years. The overall injury rate was 5.2 per 100 worker years, but housekeepers had the highest rate at 7.9, and they also had the highest rate of musculoskeletal disorders.

A home economics class practises bed making in New South Wales, around 1937. Photo: State Records NSW via Wikimedia Commons (no known copyright restrictions)

Robots came to cloth slowly, through research labs. In 2018 and 2019 a team led by Ken Goldberg at UC Berkeley and industry partners published a bed making study using Toyota's HSR and a Fetch mobile manipulator. A neural network looked at depth images and picked the best corner of the blanket to grab and pull. On a quarter scale bed the robots reached 92 percent blanket coverage over 193 trials, against 95 percent for a human. Inventors also tried to skip the robot: from 2012, several self-making beds that pull their own covers straight were in development.

A hand of the ARMAR humanoid at the Karlsruhe Institute of Technology grasps household goods in a lab kitchen, 2013. Photo: NatiSythen / Wikimedia Commons (CC BY-SA 3.0)

Humanoids making beds in real homes in 2026

On 17 September 2026 Figure published results for its Helix 2.5 control policy, tested in 30 Bay Area homes the robot had never seen. Bed making was its strongest task, with 94 complete successes out of 140 tries, or 67 percent. Towel folding came in at 62 percent. Figure scored no partial credit, and any safety intervention counted as a failure. Two out of three is good for cloth, and it still leaves the bed unmade one morning in three. For more on that gap, see our post on robots in unseen homes.

A Unitree G1 humanoid, one of the general purpose robots now being trained on household chores. Photo: Sayanesy / Wikimedia Commons (CC0)

Some companies sell the chore by the hour. Tau Robotics in San Francisco charges 30 dollars an hour for a humanoid that cleans homes, and a customer's video of it making a bed went viral in August. USA Today reported that a remote human operator was directing the work, and the CEO told CBS News that people wearing VR goggles still guide the robots. Around the same week, DYNA Robotics showed Taku, a two-armed robot on wheels that loads washers, folds towels and shelves them for hotels and laundromats, and Flourish Robots launched a 3,555 dollar home robot it says can learn a new chore in 30 minutes.

A rumpled hotel bed waiting to be made, the starting state for robot bed making trials. Photo: Kurt Kaiser / Wikimedia Commons (CC0)

ABB does not sell a bed making robot, but its YuMi IRB 14000 shows up often in cloth research. YuMi has two seven-axis arms, each rated for 0.5 kg, and it was designed to work beside people without a fence. In 2022 Berkeley and Karlsruhe researchers built SpeedFolding on a physical ABB YuMi with parallel grippers. It smoothed and folded crumpled shirts in under two minutes on average, reaching 30 to 40 folds per hour with a 93 percent success rate. Our earlier post on laundry and cloth handling covers more of this two-armed cloth work.

ABB's dual arm YuMi IRB 14000, the robot used in the SpeedFolding research, working beside a lab worker. Photo: ABB

Where bed making robots go next, at home and in hotels

Hotels are the more likely first buyers. Their beds are the same size in every room, the linen is standard, and the job is repeated hundreds of times a day. A realistic early step is a robot that helps the attendant by carrying linen, lifting the mattress edge or folding sheets in the laundry, while the person still makes the bed. Two-armed robots like Taku already target that back-of-house work. ABB has shown a mobile YuMi concept on a wheeled base for hospital labs. Putting a robot like that in a hotel corridor would be a new use, and for now it is only a possibility.

ABB's mobile YuMi concept, a dual arm robot on a wheeled base, in a hospital lab corridor. Photo: ABB

Payload is the first hard limit. A duvet or a stack of sheets is heavier than YuMi's 0.5 kg per arm, so a full bed would need something closer to ABB's GoFa CRB 15000, a cobot rated for 5 kg, or a humanoid with stronger arms. Grip is the second. Cloth folds over itself, hides its corners and changes shape every time it is touched, which is why the Berkeley teams trained on thousands of images. Any such cell would be planned and tested in simulation first, the way industrial cells are built in RobotStudio, though simulating cloth is still much weaker than simulating rigid parts.

An ABB GoFa CRB 15000 cobot at a work table. Photo: ABB

At home, the pace depends on price and on whether people trust a machine next to the bed. A home robot that leans over a bed works right where people sleep, and children and pets may be in the room. Most 2026 services still rely on remote operators, which raises privacy questions about cameras in bedrooms. Prices have dropped from tens of thousands of dollars toward a few thousand, yet a 67 percent success rate is still far below what a person does without thinking. The likely path is hotels and care homes first, with helper robots that fetch and fold, and fully autonomous home bed making robots a few years behind.

An ARMAR humanoid robot in the home-like lab at the Karlsruhe Institute of Technology, 2013. Photo: NatiSythen / Wikimedia Commons (CC BY-SA 3.0)

Innovation

AI solutions for effortless ABB robot control.

Automation

Robotics

ceojohntran@chatwithrobot.net

+84905311611

© 2025. All rights reserved.

qtran1215@gmail.com