BRENNEN

Industries / Government, education and community / Childcare

Daily care and safety

Daily care and safety is the hands-on work of keeping young children fed, clean, rested and unhurt while they are away from their parents, and the rules that say how that work is done.

7

Milestones achieved, from 1981 to 2023.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1981

    Handwashing halves diarrhea

    Diarrhea spreads easily among children in day care. Researchers ran a handwashing program in two US day care centers and compared them with two centers without one. After the second month the program centers always had less diarrhea, and over the 35 weeks of the study their rate was about half that of the other centers.

    eutils.ncbi.nlm.nih.gov

    Credit

    R. E. Black, E. J. Gangarosa and colleagues

  2. 2000

    Staff training tested by lottery

    In a cluster randomized trial, in which whole centers rather than single children were assigned to the program or not by chance, Australian researchers trained child care staff in how infections spread and in handwashing for staff and children. Children in the trained centers had 1.9 bouts of diarrhea a year against 2.7 in the others, a 50% reduction after adjustment, and 66% in centers where children washed their hands most reliably. The effect was limited to children older than 24 months.

    eutils.ncbi.nlm.nih.gov

    Credit

    L. Roberts and colleagues, Australian National University

  3. 2000

    Infant deaths in child care

    Sudden infant death syndrome (SIDS) is the unexplained death of a baby, usually during sleep. In 1,916 SIDS cases from 11 US states between 1995 and mid-1997, 20.4% happened in child care, where census figures suggested about 7%. Babies who died in child care were more likely to have been put down or found on their stomachs when they usually slept on their backs or sides. A follow-up of 2001 deaths in 13 states found 16.5% (79 of 480) in child care, and no longer any difference in sleep position from deaths at home.

    Lessonthe authors of the follow-up concluded that teaching child care providers to put babies on their backs had worked, and that the teaching should reach unregulated carers too.

    eutils.ncbi.nlm.nih.goveutils.ncbi.nlm.nih.gov

    Credit

    R. Y. Moon and colleagues, Children's National Medical Center, Washington, DC

  4. 2012

    Crèches prevent drownings

    Living conditions in rural Bangladesh expose small children to drowning. From 2006 to 2010 the Anchal program kept 18,596 young children in crèches, small village day care centers with local carers. Compared with matched children who did not attend, their risk of dying by drowning was 0.181 times as high, at a cost of USD 50.74 to 60.50 per child per year. A 2022 process evaluation counted 400 centers averaging 22.2 children each.

    Lessonchildren aged one to two, the most at risk of drowning, were the least likely to enroll and attend, because of low interest in the activities and parents' safety concerns.

    eutils.ncbi.nlm.nih.goveutils.ncbi.nlm.nih.gov

    Credit

    Centre for Injury Prevention and Research, Bangladesh (F. Rahman, A. Rahman and colleagues); process evaluation by The George Institute for Global Health

  5. 2015

    A camera in every room

    In January 2015 a teacher at a day care center in Songdo, Incheon, punched a 4-year-old in the chest and threw her to the ground. South Korea's revised infant care law, approved in May 2015 and in force from 19 September 2015, required at least one closed-circuit television (CCTV) camera in every room of the country's more than 40,000 day care centers, footage kept for 60 days, and parents allowed to view it.

    Lessonreported abuse in day care centers rose from 432 cases in 2015 to 1,371 in 2019, by the count of Korea's Child Rights Protection Agency, and many later cases were caught on camera; Gong Hyejeong of the Korea Child Abuse Prevention Association said cameras do not stop abuse on their own and that higher awareness explains part of the rise.

    koreatimes.co.krasiae.co.kr

    Credit

    South Korea's National Assembly and Ministry of Health and Welfare

  6. 2017

    Day care meal rules rewritten

    The US Child and Adult Care Food Program (CACFP) pays day care centers and family day care homes back for meals that meet set food rules. A rule published on 25 April 2016, with compliance from 1 October 2017, made the first major change to those meal patterns since the program began in 1968: juice at most once a day, breakfast cereals with no more than 6 grams of sugar per dry ounce, more whole grains and vegetables, support for breastfeeding, and no deep-fat frying on site.

    federalregister.gov

    Credit

    US Department of Agriculture, Food and Nutrition Service

  7. 2023

    Alarms on nursery buses

    On 5 September 2022 a girl died after being left on the shuttle bus of a certified center for early education and care (a kodomoen) in Makinohara, Shizuoka, Japan. The Ministry of Land, Infrastructure, Transport and Tourism (MLIT) set up a working group to write specifications for safety devices. From April 2023, checking where every child is and fitting such devices became mandatory, on about 44,000 buses by the count of NHK, Japan's public broadcaster. The system NHK described sounds an alarm when the engine stops until the driver presses a button at the back of the bus, and can text staff if its camera or microphone detects a child.

    Lessona survey of kindergartens and certified centers afterwards found that driver training and seat belts, neither of them required, still lagged behind.

    mlit.go.jpjstage.jst.go.jpbusworldeurope.org

    Credit

    Government of Japan, MLIT

Watch for

  1. CCTV trial in 300 centers

    Australia's federal, state and territory education ministers agreed in August 2025 to an AUD 21 million trial of CCTV cameras in up to 300 small and medium child care centers, part of an AUD 189 million safety package, to start in October or November 2025. The trial is to test camera placement, who owns and stores the footage, and how to keep it secure, and is to be designed and reviewed by an outside organization with the Australian Centre for Child Protection and the national regulator, ACECQA.

    Achieved whenthe outside review is published and reports what the cameras caught or deterred, what they missed, and whether any footage was misused or leaked.

    aussiechildcarenetwork.com.auaapnews.aap.com.au

    Claimed by

    Australian federal, state and territory education ministers, August 2025

Open

  1. 01

    Test bus left-behind alarms

    A public bench test of the devices sold to Japanese nurseries and kindergartens that detect a child left on a bus. Weighted infant and toddler manikins, warmed with low-power heating pads and given a small breathing motion by a servo, are hidden in different seats, on the floor and under blankets in a parked, shaded minibus with the engine off, and each device's alerts are timed. No child takes part.

    The brief
    Credit · unclaimed
  2. 02

    Staged safe-sleep photo test

    A public set of at least 1,000 photos of weighted infant dolls in cribs, play yards, adult beds, sofas, car seats and inclined products, staged by volunteers in real rooms, each labeled against the American Academy of Pediatrics' 2022 safe sleep recommendations (on the back, on a firm and flat surface that is not inclined, no soft bedding, no bed sharing), and a script that scores how well general-purpose vision AI models, the kind built into phones and chat apps, spot each hazard. No photo shows a real child.

    The brief
    Credit · unclaimed

Open 01 · the brief

Test bus left-behind alarms

Credit · unclaimed

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

What to build

A public bench test of the devices sold to Japanese nurseries and kindergartens that detect a child left on a bus. Weighted infant and toddler manikins, warmed with low-power heating pads and given a small breathing motion by a servo, are hidden in different seats, on the floor and under blankets in a parked, shaded minibus with the engine off, and each device's alerts are timed. No child takes part.

Why it is still open

MLIT publishes a list of devices by maker and model, but it gives product information, not test results; searches in Japanese and English found no published independent comparison of how often these devices find a hidden child or raise false alarms. A survey in the Japanese Journal of Safety Education reports on how centers run their buses, not on device performance.

mlit.go.jpjstage.jst.go.jp

Built with what exists

a used microbus or van; three to five listed devices of both kinds (camera, radar or infrared sensors, and driver-button alarms); infant and toddler CPR training manikins; heating pads; a hobby servo and microcontroller for breathing motion; a time-stamped camera.

  1. First step

    Buy one automatic-detection device and one infant manikin and run 20 placements in a parked van to fix the method.

  2. Done looks like

    For each device, 50 placements over 10 hiding spots in light, dark and under-blanket conditions, with the share detected and the median time to an alarm outside the bus, plus false alarms over 20 runs with the bus empty; all videos and data published.

Open 02 · the brief

Staged safe-sleep photo test

Credit · unclaimed

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

What to build

A public set of at least 1,000 photos of weighted infant dolls in cribs, play yards, adult beds, sofas, car seats and inclined products, staged by volunteers in real rooms, each labeled against the American Academy of Pediatrics' 2022 safe sleep recommendations (on the back, on a firm and flat surface that is not inclined, no soft bedding, no bed sharing), and a script that scores how well general-purpose vision AI models, the kind built into phones and chat apps, spot each hazard. No photo shows a real child.

Why it is still open

Earlier image sets either use photos of real infants or cover part of the list. A University of Florida team labeled 606 smartphone photos of infants for five safe-sleep areas (location, surface, position, soft items and nearby hazards) and reached 81% accuracy with a model built for the job. Northeastern University's CribHD set (2024) covers toys and blankets in cribs, and SMaL (2022) records an infant mannequin's pose under covers in depth and pressure images. None scores general-purpose vision models against the full list of hazards, and none stages adult beds, sofas, car seats and inclined products with dolls only.

faculty.eng.ufl.edugithub.comarxiv.orgeutils.ncbi.nlm.nih.gov

Built with what exists

weighted infant dolls; phone cameras; Label Studio for labels; open-weight vision-language models run on a local computer with llama.cpp, which is built on ggml (see the Note "A first stable number for local AI"); small models adapted on one graphics chip, such as Liquid AI's LFM3 (see the Note "One GPU to fine-tune").

  1. First step

    Turn the AAP recommendations into a label sheet and stage 100 photos in three rooms with two dolls.

  2. Done looks like

    1,000 photos from at least 20 rooms, each labeled by two people with their agreement reported, and the share of each hazard found and the false-alarm rate for at least five models; photos, labels and code published.