BRENNEN

Industries / Construction and property / Real estate / Short-term rentals

Cleaning between stays

This is the turnover: the cleaning, laundry, restocking and checking done between one guest's checkout and the next guest's arrival, usually by cleaners the host hires.

4

Milestones achieved, from 2014 to 2022.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2014

    Photo checklists for cleaners

    Properly, a San Francisco company started by hosts in 2014, built its first product around checklists that ask the cleaner for a photo of each finished step, so a host who is not there can see the work. It later added automatic scheduling, a marketplace of cleaners and training, and says it has managed more than 4.5 million turnovers for over 10,000 hosts in more than 45 countries (its own figures).

    getproperly.com

    Credit

    Properly, Inc

  2. 2016

    Cleanings booked by software

    TurnoverBnB, founded in 2016, sells hosts software that schedules cleanings automatically, pays the cleaner and shares checklists and photos, and runs local marketplaces of independent cleaners in the US, Canada, Europe and Australia. It renamed itself Turno in February 2023.

    newswire.com

    Credit

    Assaf Karmon, co-founder and chief executive, and the TurnoverBnB team

  3. 2020

    A cleaning standard for homes

    On 27 April 2020 Airbnb announced a cleaning protocol for hosts built on US Centers for Disease Control and Prevention (CDC) guidance, with advice from former US Surgeon General Vivek Murthy and the hygiene company Ecolab. It called for masks and gloves, approved disinfectants and a 24-hour wait before entering after a guest left; hosts who could not commit could block 72 hours between stays instead. Airbnb called it "the first overarching standardized guidelines for cleaning and sanitization in the home sharing industry", and from 20 November 2020 required every host outside mainland China to commit to its five-step process or face warnings, suspension or removal.

    news.airbnb.comnews.airbnb.comnews.airbnb.com

    Credit

    Airbnb; Vivek Murthy; Ecolab

  4. 2022

    Limits on checkout chores

    In November 2022 Airbnb said guests "should not have to do unreasonable checkout tasks", such as stripping beds, doing laundry or vacuuming, while turning off lights, putting food in the bin and locking the door stayed reasonable; hosts' checkout requests had to be shown before booking. In October 2026 Airbnb told hosts that from 20 November it would generally not allow requests to wash dishes, strip beds or take out the rubbish, and that repeated breaches could bring a notice on the listing or lower it in search.

    LessonWhen guests pay a cleaning fee, the split of cleaning work between host and guest becomes a platform rule, not each host's choice.

    fox5dc.comskift.com

    Credit

    Airbnb

Watch for

  1. Humanoids make beds in homes

    Figure says its Helix 2.5 robot model, trained first on its own large collection of recorded human activity, made beds, folded towels and tidied living rooms in 30 Bay Area homes it had never seen. A bed counted as made when both pillows and the comforter's corners were at the head of the bed and the comforter was pulled smooth. Across the three tasks, with no partial credit, it succeeded in 56% of trials, against 9% for the same model trained without the human data. In May 2026 two of its humanoids reset a bedroom, working together to make the bed, in under two minutes.

    Achieved whenan outside party grades the robots on beds and rooms it chooses and publishes success rates for each task.

    figure.aifigure.ai

    Claimed by

    Figure AI, September 2026

  2. AI checks turnover photos

    Properly sells a $5 inspection in which AI flags likely problems in the cleaner's checklist photos and a person checks every flag before the next guest arrives; it says that with its checklists and photo checks "most hosts eliminate 4 out of 5 turnover errors", without giving a sample or method. RapidEye, which compares each turnover's photos with earlier photos of the same home, reports 132 issues per 100 inspections across 20,385 inspections of 2,503 holiday homes from August to October 2026, 59% of them missed cleaning and 24% damage. It notes that the companies chose to use it, so they are not a random sample, and that it does not know how many of those issues a person would also have found.

    Achieved whena vendor publishes how many real faults its system catches and how many false alarms it raises on photos it did not choose, or an outside tester does.

    getproperly.comrapideyeinspections.com

    Claimed by

    Properly, on a product page read in November 2026; RapidEye, November 2026

Open

  1. 01

    Robot linen change test

    A robot that strips a used bed and remakes it with clean linen: fitted sheet on the mattress, pillows in pillowcases, duvet in its cover. It is graded step by step against a published checklist.

    The brief
    Credit · unclaimed
  2. 02

    Public turnover photo test

    An open set of turnover photos, taken in rooms the builders stage themselves, each labelled against a standard checklist (bed made with clean linen, towels set out, bin emptied, no stains, nothing missing, nothing broken), with known faults planted, and a scoring script anyone can run on an AI checker.

    The brief
    Credit · unclaimed

Open 01 · the brief

Robot linen change test

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A robot that strips a used bed and remakes it with clean linen: fitted sheet on the mattress, pillows in pillowcases, duvet in its cover. It is graded step by step against a published checklist.

Why it is still open

Published robot results stop before the linen is changed. A 2019 Berkeley and Toyota study had robots pull a blanket over a quarter-scale bed, reaching 92% coverage over 193 trials, and Figure's 2026 humanoids straighten a comforter and pillows that are already on the bed. Searches of the web, of Hugging Face datasets for "duvet", "fitted sheet" and "bed making", and of GitHub found no robot that puts on a fitted sheet or a duvet cover.

arxiv.orgfigure.ai

Built with what exists

A pair of low-cost robot arms such as the SO-101 kit used with Hugging Face's LeRobot software (Notes: "No download"); robot models that learn a task from a few demonstrations, such as NVIDIA's GR00T 2 (Notes: "Five demonstrations per task") or Runway's open-weight Praxis-1 (Notes: "Open robot model weights"); a crib or cot mattress with standard fitted sheets to start, and overhead cameras.

  1. First step

    Record 50 human-guided demonstrations of putting a fitted sheet on a crib mattress with two LeRobot arms and publish them as an open dataset.

  2. Done looks like

    Twenty filmed attempts from a stripped mattress, with a published tally of each step done to the checklist (four fitted-sheet corners on, each pillowcase on, duvet fully inside its cover, no bunching), the time taken, and the same tally for a person doing the same bed.

Open 02 · the brief

Public turnover photo test

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An open set of turnover photos, taken in rooms the builders stage themselves, each labelled against a standard checklist (bed made with clean linen, towels set out, bin emptied, no stains, nothing missing, nothing broken), with known faults planted, and a scoring script anyone can run on an AI checker.

Why it is still open

The companies selling AI turnover checks publish no catch rates or false-alarm rates. RapidEye, which publishes the most detail, counts the issues its system reported in 20,385 inspections and says it does not know how many of them a person would also have found. The only related open benchmark found, HousekeepingOps-3D on Hugging Face, uses simulated hotel rooms rather than photos of real ones; searches of Hugging Face datasets for "housekeeping", "room cleanliness" and "turnover inspection" found nothing else.

rapideyeinspections.comhuggingface.co

Built with what exists

Phones; a team member's own flat or a hospitality school's practice room, with no guests or personal belongings in frame; Label Studio for labelling; vision AI models such as Claude Fable 5.2 (Notes: "Three working devices") and Gemini Decision 1, which answers many yes-or-no checklist questions about an image in one pass (Notes: "A thousand questions, one pass").

  1. First step

    Write a 30-item turnover checklist, stage one bedroom and one bathroom in 15 fault conditions each, and photograph them from the angles a cleaner's app asks for.

  2. Done looks like

    At least 1,000 photos from at least 20 rooms, with at least 300 planted faults of 10 kinds, released under an open licence, and the share of faults caught and the false-alarm rate reported for at least three AI models.