BRENNEN

Industries / Health / Healthcare

Home health

Home health is nursing, therapy and personal care, such as help with bathing, dressing and medicines, given to people in their own homes rather than in a hospital or care home.

1859

District nursing began in England in 1859, when the Liverpool philanthropist William Rathbone asked Mary Robinson, the nurse who had cared for his dying wife at home, to nurse the city's poor in their own homes. He later worked with Florence Nightingale to train nurses for the city's 18 districts.

qicn.org.uk

39

Milestones achieved across 6 domains.

10

To watch for: announced, not yet confirmed outside.

12

Open deliverables. Nobody has shipped them.

In one minute

Where it started

Organized home nursing began in Liverpool in 1859 and spread through charities and visiting nurse associations. It became a large industry when governments began to pay for it: the US through Medicare from 1966, Germany through long-term care insurance from 1995 and Japan from 2000. Each payer then wrote rules for how visits are assessed, recorded and billed.

Where it stands

In OECD countries an average of 69% of people who receive formal long-term care get it at home, from 34% in Portugal to 95% in Israel, by the OECD's count in Health at a Glance 2023. The US Bureau of Labor Statistics counts 4,677,100 home health and personal care aide jobs in 2025, at a median pay of $17.21 an hour. Japan's health ministry estimates it needs about 2.4 million care workers in fiscal year 2026 against 2.15 million in 2022. Visits are planned by software, checked in by phone and paid by detailed rules, but the work itself is still done by hand, one home at a time.

What is open

Much of what is sold to home care agencies, from scheduling agents and radar fall sensors to AI note-taking, rests on the seller's own numbers. There are no public, repeatable tests of consumer fall detectors, no shared multi-day benchmark for visit planning, and no open checkers for Medicare's paperwork rules. Cheap robot arms, open vision models and public payment data now let a small team build and publish these tests without touching a real patient.

Domains

AchievedWatch forOpen

Advances to build with

1 research note

Open in home health

3 of 12 open deliverables

Open · Scheduling and routing visits

A multi-day planning benchmark

A public set of synthetic home-visit planning problems that cover 5 to 7 days instead of one, scored on how often each patient sees the same caregiver (continuity of care) as well as on travel time and lateness, with real road travel times in at least three countries outside Italy.

Built with what exists

the EU GEOSTAT population grid or WorldPop for where people live; OpenStreetMap and OSRM for travel times; the Udine JSON format and Python validator, extended to several days; Google OR-Tools, Timefold or PyVRP as baseline solvers.

Credit · unclaimed

Open · Nursing and therapy visits

Robot arm fetches and opens

A control policy for a low-cost SO-101 robot arm that, in a staged kitchen, picks up a named object (a cup, glasses, an empty pill bottle) and places it in the open hand of a mannequin, and unscrews the plain cap of an empty bottle.

Built with what exists

an SO-101 arm (about $120–150 in parts) and Hugging Face's LeRobot software (see the Note "No download"); robot models that learn from a few demonstrations, such as NVIDIA GR00T 2 (see the Note "Five demonstrations per task") or Runway Praxis-1 (see the Note "Open robot model weights"); two USB cameras.

Credit · unclaimed

Open · Remote monitoring

Bench test of fall detectors

A published, repeatable test of five or more consumer fall detectors, such as recent Apple, Samsung and Google watches and two alarm pendants. Professional stunt performers fall onto crash mats in scripted ways (forward, backward, sideways, slow slide, fall from a chair), with video as the record of what happened, and volunteers wear each device in daily life to count false alarms per day.

Built with what exists

retail devices; gymnastics crash mats; two video cameras; the measures the 2018 review recommends (sensitivity, precision, false alarms per hour of wear); a speech-to-text model such as ElevenLabs Scribe v3 to log spoken trial notes (see the Note "Fewer word errors").

Credit · unclaimed

Where it sits