BRENNEN

Industries / Consumer goods / Baby products

Car seats

Car seat work is the design, testing and fitting of child restraints for cars, from rear-facing infant carriers and convertible seats to boosters, and of the anchors and rules that hold them in the car.

7

Milestones achieved, from 1964 to 2013.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1964

    Seats turned to face backward

    Early child seats kept a child in place rather than protecting it. The Henry Ford museum says a seat from around 1930 "offered little protection during an accident", and London's Science Museum describes a European seat of about 1958 to 1963 as "intended as a restraint rather than a safety device". In 1964 Bertil Aldman of Chalmers University of Technology in Sweden presented the first rear-facing child seat prototype, inspired by the way astronauts lie on their backs at launch, and it went into production in 1967, by the account of a review led by Katarina Bohman of Volvo Cars. Child seat use in Sweden rose from 22% in 1976 to 95% in 1990, before a 1998 regulation first required child seats. The same review reports that 83% of Swedish children up to age three still ride rear-facing, and that 2021 was the country's first year with no child killed as a car passenger.

    Lessonin Sweden authorities, makers, insurers and consumer groups gave the same advice, and use became near universal before any law required it.

    thehenryford.orgcollection.sciencemuseumgroup.org.ukstatic.nhtsa.gov

    Credit

    Bertil Aldman, Chalmers University of Technology

  2. 1978

    First law requiring child seats

    Tennessee's Child Passenger Protection Act, passed in 1977 after lobbying by the pediatrician Robert Sanders and his wife Pat, came into force on 1 January 1978. A 1988 paper by Michael Decker and colleagues calls it "the nation's first law requiring the use of child restraint devices". By the law's third year, A. F. Williams and J. K. Wells counted the share of Tennessee children in restraints anchored by seat belts rising from 8% before the law to 29%, against 11% to 14% in Kentucky, which had no law. Travel in a parent's arms, which the law allowed, had not changed. Within seven years every US state had passed a similar law, by Vanderbilt University Medical Center's account.

    Lessonan exemption for children held in arms left a hazardous practice legal and untouched.

    eutils.ncbi.nlm.nih.goveutils.ncbi.nlm.nih.goveutils.ncbi.nlm.nih.govnews.vumc.org

    Credit

    Robert and Pat Sanders; Tennessee General Assembly

  3. 1981

    Crash tests for US seats

    The US National Highway Traffic Safety Administration (NHTSA) adopted the first federal standard for child seats, FMVSS 213, in 1971. It required a seat belt to hold the seat in the car and a harness to hold the child, but it had no crash test and did not cover rear-facing infant seats. A stricter version in force from 1 January 1981 added a frontal crash test at 30 mph (48 km/h) and a minimum force to open the buckle, so that children cannot release the harness, and it covered rear-facing infant seats, car beds and forward-facing seats for children under 50 lb (23 kg).

    Lessonfor its first ten years the US standard let child seats be sold without being crash tested.

    saferidenews.com

    Credit

    NHTSA

  4. 1996

    Airbags kill children in front

    The US Centers for Disease Control and Prevention (CDC) counted 32 children killed by passenger airbags from January 1993 to November 1996. Twenty-one were unrestrained or wrongly restrained, and nine were in rear-facing child seats placed in the front passenger seat. CDC's advice became that all children aged 12 and under be properly secured in the back seat.

    Lessona device designed to protect adults killed infants whose seats were placed in front of it, and the back seat became the rule for children.

    cdc.gov

    Credit

    CDC; NHTSA

  5. 1999

    Anchors built into cars

    In 1994 Dana Hutchinson, aged three, died in her child seat because the car's front seat belt could not hold the seat without a special add-on buckle. NHTSA formed a Blue Ribbon Panel on child seat and vehicle compatibility, which in 1995 called for a universal anchorage that does not rely on seat belts, first called ISOFIX after the design of a working group of the International Organization for Standardization (ISO). NHTSA's rule of March 1999, FMVSS 225, cited a 1996 study in four states in which about 80% of people made at least one significant error using child seats. It required top tether anchorages (a fixing point for a strap from the top of the seat) in most passenger cars from 1 September 1999, and lower anchorages in all new vehicles, with matching attachments on child seats, from 1 September 2002; the US system is known as LATCH. Australia's 1970 standard, AS E46, had already required at least three points of attachment between child seat and car, and by 1998 fewer than 5% of child seats in New South Wales were used without a top tether. NHTSA's 2011 survey of 4,167 children's seats at 24 sites still found 46% of car seats and boosters misused, and 45% of drivers who felt confident about their installation had made at least one error.

    Lessonbuilt-in anchors removed one source of error, but about half of all seats were still misused a decade later.

    saferidenews.comfederalregister.govstatic.nhtsa.govcrashstats.nhtsa.dot.gov

    Credit

    NHTSA and its Blue Ribbon Panel; the ISO working group

  6. 2008

    Boosters rated for belt fit

    A booster raises a child so that the adult lap and shoulder belt fits; the Bohman review credits Sweden with the world's first belt-positioning booster, in 1978. On 1 October 2008 the US Insurance Institute for Highway Safety (IIHS), with the University of Michigan Transportation Research Institute, published ratings of how well 41 boosters placed the belt on a child: 10 were "best bets" and 13 were not recommended at all. "We evaluated the safety belt fit boosters provide, not crash protection," said IIHS president Adrian Lund. IIHS rated boosters until 2024, and by the end most boosters on the market earned its top rating. A December 2023 NHTSA rule raised the minimum weight at which makers may recommend a booster from 13.6 kg (30 lb) to 18.4 kg (40 lb).

    Lessona published test of belt fit alone, with no crash test, moved most boosters on sale to the top rating.

    insurancejournal.comiihs.orgfederalregister.gov

    Credit

    IIHS; University of Michigan Transportation Research Institute; NHTSA

  7. 2013

    Height sizing and side tests

    UN Regulation 129, known as i-Size, entered into force on 9 July 2013 under the UN's 1958 agreement on shared vehicle rules. It sizes seats by the child's height rather than weight alone, keeps children rear-facing up to 15 months and adds side impact tests. Since 1 September 2024 the EU has allowed only seats approved under it to be placed on the market, by the account of the Spanish child road safety alliance AESVi, though no date has been set to stop using older seats. The US adopted its own side impact test, FMVSS 213a, in June 2022, ten years after Congress ordered one. Seats for children up to 18.1 kg (40 lb) are tested on a sled that simulates a side crash at about 30.5 km/h (19 mph) against an intruding door, and NHTSA estimated it would prevent 3.7 deaths and 41 injuries a year. Compliance was due on 30 June 2025, but NHTSA stopped enforcing it from that day after makers said demand for test facilities limited their access, and a September 2026 rule set the date at 5 December 2026. China replaced its 2011 national standard for child restraints with GB 27887-2024 from 1 July 2025.

    Lessona new crash test needs laboratories able to run it, and the US side test will be enforced more than four years after it was adopted.

    legislation.gov.ukinterempresas.netfederalregister.govfederalregister.govfederalregister.govcodeofchina.com

    Credit

    the parties to the UN 1958 Agreement; NHTSA; China's State Administration for Market Regulation (SAMR)

Watch for

  1. Seats that check their install

    Thule says its Connected Car Seats, an ISOFIX base with an infant seat and two toddler seats, use sensors to give "instant feedback on the most important safety steps", including that the seat and base are installed correctly and that the harness is tightened on the child. Alerts, including an abandoned child alert, appear on the phone or on the car's screen through Apple CarPlay and Android Auto. The seats are to be sold in selected European markets from September 2026. The ADAC test wins Thule cites are for the seats, not the sensors.

    Achieved whenan independent tester such as ADAC or Stiftung Warentest stages common installation and harness errors with test dummies or weighted dolls and publishes how many the sensors catch and how many false alerts they raise.

    thule.comnurserytoday.co.uk

    Claimed by

    Thule, September 2026

  2. Easier anchors, later

    NHTSA's January 2025 rule requires each lower anchorage in a car to sit no more than 25 mm inside the seat bight (the gap between seat back and cushion), with a clearance angle of at least 54 degrees for the child seat's connector, and standard ISO symbols to mark the anchorages. It was to reach 20% of new vehicles from September 2028 and all of them after August 2030. In August 2026, answering petitions for reconsideration, NHTSA proposed stretching the phase-in so that every vehicle complies from 1 September 2031. NHTSA did not count the benefits itself; it cites an earlier estimate of about 3 lives and 6 serious injuries saved a year if correct installation and tether use each rose by 5%.

    Achieved whena final rule fixes the dates, and a field survey, such as NHTSA's next misuse study, shows more correct installations in vehicles that meet the rule.

    federalregister.govfederalregister.gov

    Claimed by

    NHTSA, August 2026

Open

  1. 01

    Photo check of seat installs

    A public set of at least 2,000 photos and short videos of car seats installed in real cars, each holding a weighted infant or toddler doll, staged by volunteers and checked by certified child passenger safety technicians. Each is labeled for the errors NHTSA's survey defined: a rear-facing seat for a baby under one reclined less than 30 degrees, a seat that moves more than 2 inches (5 cm) when pushed or pulled at the belt path, a booster's lap belt across the abdomen, plus harness slack, chest clip height and whether the top tether is attached. A script then scores how well vision AI models, the kind built into phones and chat apps, find each error. No child is photographed.

    The brief
    Credit · unclaimed
  2. 02

    Recall check from the label

    A phone tool and a public test set. The tool reads a car seat's label from a photo (maker, model number, date of manufacture), matches it against NHTSA's recall file, and reports any recall that covers that model and date, along with the use-by period stated in the maker's manual where there is one. The test set holds at least 500 label photos from seats in shops, from owners, and from second-hand sellers who agree, each with the right answer checked by hand. No child appears in any photo.

    The brief
    Credit · unclaimed

Open 01 · the brief

Photo check of seat installs

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 2,000 photos and short videos of car seats installed in real cars, each holding a weighted infant or toddler doll, staged by volunteers and checked by certified child passenger safety technicians. Each is labeled for the errors NHTSA's survey defined: a rear-facing seat for a baby under one reclined less than 30 degrees, a seat that moves more than 2 inches (5 cm) when pushed or pulled at the belt path, a booster's lap belt across the abdomen, plus harness slack, chest clip height and whether the top tether is attached. A script then scores how well vision AI models, the kind built into phones and chat apps, find each error. No child is photographed.

Why it is still open

NHTSA's 2011 survey found 46% of seats and boosters misused. Nearby work stops short. SVIRO, a synthetic image set of the rear seats of ten vehicles, is built to classify what occupies each seat, not whether a child seat is installed correctly. A commercial classifier from Nyckel names the brand of a seat in a photo. Evenflo's SensorSafe chest clip warns when the clip is unbuckled while driving, the back seat is too hot or cold, or a child is left buckled in, but does not check the installation. Thule's sensor seats check only Thule's own models and have no outside test. Searches in November 2026 found no tool that judges a car seat installation from a photo.

crashstats.nhtsa.dot.govarxiv.orgnyckel.comevenflo.comthule.com

Built with what exists

weighted infant and toddler dolls or CPR training manikins; phone cameras; a digital angle gauge for recline; a push-pull force gauge and a ruler, filmed, for movement at the belt path; 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"); Gemini Decision 1 to score many yes/no checklist questions about each photo in one pass (see the Note "A thousand questions, one pass"); a small model adapted on one graphics chip, such as Liquid AI's LFM3 (see the Note "One GPU to fine-tune").

  1. First step

    Turn NHTSA's misuse definitions into a label sheet, then stage 100 installations of two seats in three cars with one doll.

  2. Done looks like

    2,000 labeled installations across at least 30 vehicle models and 20 seat models; every one labeled by two technicians, with agreement (Cohen's kappa, a score that corrects for chance agreement) of at least 0.7; for at least five models, the share of each error found and the false-alarm rate on correct installations; a published list of the errors a photo cannot show; photos, labels and code released.

Open 02 · the brief

Recall check from the label

Credit · unclaimed

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

What to build

A phone tool and a public test set. The tool reads a car seat's label from a photo (maker, model number, date of manufacture), matches it against NHTSA's recall file, and reports any recall that covers that model and date, along with the use-by period stated in the maker's manual where there is one. The test set holds at least 500 label photos from seats in shops, from owners, and from second-hand sellers who agree, each with the right answer checked by hand. No child appears in any photo.

Why it is still open

NHTSA estimates that 38 makers sell 16 million child seats a year in the US and that 17.72% are registered with the maker, so a recall letter reaches fewer than one owner in five directly. NHTSA's recall file marks child restraint recalls and gives the first and last dates of manufacture covered, which is what a label match needs. US labels must show the model name or number and the month and year of manufacture, and Canadian labels the model name, number and date of manufacture; neither rule asks for an expiry date, so use-by periods must be gathered from makers' manuals one by one. Searches in November 2026 for an app that reads a car seat label from a photo found none; the only photo tool found, Nyckel's brand classifier, does not read model numbers or dates.

federalregister.govstatic.nhtsa.govecfr.govlaws-lois.justice.gc.canyckel.com

Built with what exists

NHTSA's recall flat file (tab-delimited, with a field that marks child restraint recalls); a small vision model adapted to read labels, such as Liquid AI's LFM3 (see the Note "One GPU to fine-tune"); Meta's Muse Ember to pull fields out of recall summaries and manuals (see the Note "The jobs agents repeat"); Tesseract OCR as a baseline; Browser Use 1.0 to collect manuals from makers' sites (see the Note "Deleting 80% to start over"); Python.

  1. First step

    Pull every child restraint record of the last ten years from the recall file, then photograph the labels of 30 seats in one shop and match them by hand.

  2. Done looks like

    500 label photos from at least 15 brands; model number and date of manufacture read exactly on at least 95% of photos; the recall answer agrees with a hand check on a sample of 100 photos that includes at least 20 recalled seats; a table of stated use-by periods for at least 200 models, each linked to the manual page.