Industries / Health / Healthcare / Senior living and assisted care
Dining and activities
Dining and activities is what residents eat and drink, including food softened for people who have trouble swallowing, and how they spend their days.
5
Milestones achieved, from 2002 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
2002
Eden Alternative tested
The Eden Alternative, founded by the geriatrician Bill Thomas and Jude Thomas, brings pets, plants and children into nursing homes to fight what it calls loneliness, helplessness and boredom, and says it reaches people in 22 countries. A University of Louisville team compared a home that adopted it in November 1998 with a similar home run by the same organization. After a year they found no benefit in cognition, function, survival, infections or cost, and the Eden home had more residents with recent falls and nutrition problems, and more staff leaving and being hired.
Lessonthe team's own observations of the Eden home were positive, so a change in a home's culture may take more than a year to show in measured outcomes.
Credit
Mary Coleman and colleagues, University of Louisville
2003
Screening for malnutrition
BAPEN, a British clinical nutrition association, launched the Malnutrition Universal Screening Tool ('MUST'), written by its Malnutrition Advisory Group, in 2003 and has reviewed it regularly since. In 2013 NICE, England's health guidance body, recommended BAPEN's e-learning on screening with 'MUST' for staff in hospitals, primary care and care homes.
Credit
BAPEN's Malnutrition Advisory Group
2015
One scale for soft food
People with dysphagia, trouble swallowing, are often given softened food and thickened drinks to lower the risk of choking. A survey of 2,050 people in 33 countries found 54 different names in use for food textures and 27 for drink thicknesses. The International Dysphagia Diet Standardisation Initiative (IDDSI) released a framework in November 2015 with eight numbered levels, 0 to 7, for both.
Lessondozens of local names for similar textures were replaced by one numbered scale.
Credit
the IDDSI committee, including Julie Cichero, Peter Lam and Catriona Steele
2017
Food spend measured
Quarterly surveys of 817 Australian aged care homes, covering 64,256 beds over the 2015 and 2016 financial years, found they spent AUD 6.08 per resident per day on raw food and ingredients and AUD 8.00 on all catering supplies. The authors compared this with AUD 8.25 in Australian prisons and AUD 17.25 spent by older adults living in the community, and found raw food spend fell 5% in a year while spending on supplements and food replacements rose 128%.
Credit
Cherie Hugo, Elizabeth Isenring, D. Sinclair and Ekta Agarwal
2021
Robot companion trial
The CARESSES project gave care home residents in England and Japan a Pepper humanoid robot, running conversation software designed to adapt to each person's culture, for up to 18 hours over two weeks. In a randomised trial in which 33 residents completed every step, emotional wellbeing improved compared with usual care, while the fall in loneliness was not statistically significant.
Lessona trial this small showed a gain in mood but not yet in loneliness.
Credit
Chris Papadopoulos, Antonio Sgorbissa and colleagues in the CARESSES project
Watch for
An AI companion pilot
A clinical psychologist at UNSW's Centre for Healthy Brain Ageing, Suraj Samtani, reports a 16% fall in reported loneliness among 12 residents of Leigh Place, an aged care home in southwest Sydney, after five weeks with AI companions co-designed through UNSW's Big Anxiety Research Centre. He calls it "a small pilot, not a definitive trial"; hearing and language problems made it hard for some residents, and several preferred talking with people.
Achieved whena randomised trial, as the author recommends, reports a fall in loneliness in a peer-reviewed paper.
Claimed by
Suraj Samtani, UNSW, March 2026
Light grades soft food
OptiTexture is a box costing under USD 20 that shines pulses of light from eight LEDs into a sample of softened food, records how the light scatters, adds a camera image, and assigns an IDDSI food level from 3 to 6. Its developers report 91.96% accuracy on 112 commercial dysphagia foods labelled by one rehabilitation physician, 22.32 points better than a camera alone.
Achieved whena group outside the team tests it on home-made softened meals labelled by several IDDSI-trained testers and publishes the agreement.
Claimed by
Haiyan Hu, Qian Zhang and colleagues, Hong Kong University of Science and Technology and Sun Yat-sen University, June 2026
Open
01
Purée texture by phone
A phone app that reads a video of the IDDSI fork pressure and spoon tilt tests being done on a softened food and says which IDDSI food level, 3 to 7, it meets.
The briefCredit · unclaimed02
A tray benchmark, weighed
A public set of photos of staged care-home meal trays, before and after eating, with every food weighed, and a scoreboard of how well purpose-built systems and general AI models estimate how much was eaten.
The briefCredit · unclaimed
Open 01 · the brief
Purée texture by phone
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone app that reads a video of the IDDSI fork pressure and spoon tilt tests being done on a softened food and says which IDDSI food level, 3 to 7, it meets.
Why it is still open
These tests are done by eye and hand. ViscoCam (An, Zhang and Kwong, 2021) classifies drink levels only, from the motion of a shaken cup. OptiTexture (June 2026) grades foods with its own light-and-camera box rather than the standard tests (see Watch for). The open project iddsi-open (October 2026) found that still photos cannot be labelled reliably for food levels, because the middle levels depend on size and flow that a photo does not show; its photo model rated 33% of samples at a safer level than the truth and was withheld, and its working video grader covers only the syringe flow test for drinks. No grader for filmed fork and spoon tests was found.
Built with what exists
Phone video; an open video model such as VideoMAE from Hugging Face, fine-tuned on one GPU; recipes from published texture-modified cookbooks; IDDSI's published test methods.
First step
Make 10 recipes at three textures each, film both tests, and have two testers label each sample.
Done looks like
300 filmed samples from at least 30 recipes, each labeled by two IDDSI-trained testers, and on 100 held-out samples the app agreeing with the testers at least 90% of the time and rating no more than 2% of samples at a safer level than the testers did, with a table of errors by level.
Open 02 · the brief
A tray benchmark, weighed
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of photos of staged care-home meal trays, before and after eating, with every food weighed, and a scoreboard of how well purpose-built systems and general AI models estimate how much was eaten.
Why it is still open
Purpose-built systems report their own tests. AFINI-T, from the University of Waterloo, had a mean intake error of -0.4 mL (SD 36.7 mL) on 1,039 simulated long-term care plates, by its developers' measure; DigiMeal's site gives no accuracy figures. Public weighed before-and-after sets exist for other settings: FLIC (May 2026) has 401 tray pairs from an Italian university canteen, and LeFood-Set (January 2025) has 524 pairs of Indonesian meals. None found covers care-home meals or softened and puréed food, which is the hardest to judge by eye.
Built with what exists
A kitchen scale with 1 g steps; a phone or depth camera; published care-home menus, including softened meals; Claude and other AI models that read images; a Python scoring script.
First step
Stage 30 trays from one published menu, remove varying amounts with a spoon to stand in for eating, and weigh and photograph each.
Done looks like
500 staged trays, at least 40% of them texture-modified, weighed to 1 g before and after and published under an open license, with the error of at least four methods, including staff-style estimates in quarters of a plate, reported in grams and percent.