Elderly-Care Robots: How Much of a Care Worker's Job They Actually Do

The job, and what robots would have to do

O*NET, the US Labor Department's occupation database, describes nursing assistants as people who "provide or assist with basic care or support under the direction of onsite licensed nursing staff", including "feeding, bathing, dressing, grooming, toileting, or ambulation of patients in a health or nursing facility". Their most important task is to "turn or reposition bedridden patients", rated 94 out of 100. The occupation employed 1,505,900 people in the United States in 2025, at a median wage of $20.32 an hour, and is projected to grow at an "average (3% to 4%)" rate between 2025 and 2035, according to Bureau of Labor Statistics data shown on O*NET.¹ That projection isn't a forecast about robots. The same work in older people's own homes is done by home health and personal care aides.

Most of these tasks involve lifting, turning, washing or feeding a frail person, which needs gentle force and touch. A member of China's national robot standards committee told Xinhua that robots struggle with exactly this: "turning over, feeding, supporting a person: these movements need a light touch, but robots still lack sensitive touch," and poor force control can make older people uncomfortable or unsafe.¹¹ So the robots that have reached real care work mostly do the tasks around the body: watching, reminding, carrying and keeping people company.

Japan: the best data, and it's mostly sensors

Japan has backed "care robots" for years: by 2018 its national government alone had spent "well in excess of $300 million" on research and development, according to Wright's account below,¹³ and its labour ministry publishes the clearest adoption data we found anywhere. In its survey of September to November 2025, which drew 6,876 responses, 56.4% of residential and inpatient-type care providers had care-records software and 47.2% had monitoring sensors, up from 10.2% and 30.0% in fiscal 2022. Bathing devices were in 26.2% and transfer devices in 15.8%; mobility aids, communication robots and toileting devices were each under 5%.² Among home-visit services, transfer devices were at 2.4% and monitoring at 3.1%.

The same survey shows which devices stick. Of the providers that had introduced transfer devices, 18.7% were no longer using them, and 18.8% for communication robots; for monitoring sensors it was 1.8%. Among providers using none of these technologies, 67.9% said the purchase cost was a burden.² Since fiscal 2024 the ministry has paid care providers a monthly bonus for using monitoring sensors, intercoms and records software, 100 units a month for all three or 10 for at least one, and it relaxes night staffing rules for nursing homes that meet several conditions, including monitoring sensors for every resident.³

The ministry's own field measurements, from fiscal 2020 to 2022, found that the combined time spent on direct care and night rounds fell as more beds got monitoring sensors. For wearable transfer suits, the time spent moving, transferring and repositioning residents "slightly increased" once the time to put the suit on and take it off was counted, though fewer staff reported moderate to severe back pain.³

Hug, made by FUJI, is a typical transfer device: a carer brings it to the resident and operates it to help them up from sitting for a move between bed and chair. A Wakayama Prefecture case report said it let one carer handle transfers that had needed two, including at night, but also that residents who were frightened of it sometimes moved, so a second carer had to stand behind them.⁵ In the anthropologist James Wright's fieldwork at one Japanese care home, "staff stopped using Hug after only a few days, saying it was cumbersome and time consuming to wheel from room to room." He concluded that "the care robots themselves required care".¹³

Robots that lift people on their own remain in the lab. Waseda University's AIREC, a 150 kg humanoid built with government funding, can roll a person onto their side to change a diaper or prevent bedsores, but its lead researcher "does not expect AIREC to be ready for use in nursing-care and medical facilities until around 2030", at an initial price of "no less than 10 million yen ($67,000)".¹⁴ That's the only robot we found aimed at the role's highest-rated task, and it's a prototype.

Companions at home: New York and Seoul

The largest public programmes put AI companions in the homes of older people who live alone. The New York State Office for the Aging began giving out ElliQ, a tabletop robot with a screen made by Intuition Robotics, in 2022. In August 2023 it said "more than 800 New Yorkers" had one and that ElliQ "initiates conversation, suggests activities, and remembers what users tell it", under the headline that it had achieved "a 95% reduction in loneliness".⁶ The agency's own one-page engagement report says something narrower: "95% of clients report that ElliQ is helpful in reducing loneliness and improving wellbeing", with no sample size or method given.⁶ That's self-reported helpfulness, not a measured fall in loneliness. In October 2026 the agency signed a new three-year, $2.1 million agreement, and the partners said "more than 3,500 older adults" had applied in 2025 alone and that users now interact with ElliQ 48 times a day.⁷ The robot "can remind users to take health measurements, like their blood pressure, and medications", according to The Robot Report.²⁵ ElliQ's maker says it "is not a medical device nor an emergency alert system".¹⁵

Seoul supplied 430 companion robots in 2023 and 50 more in 2024 to older people living alone, for "check-ins and medication guidance", among them Hyodol, a talking doll robot.⁸ The strongest independent study we found is mixed. In 180 low-income, socially isolated older adults in Korea, Hyodol was followed by fewer depressive symptoms and better quality of life at three months, but "these positive effects did not extend to 6 months", and the authors concluded its effects "were limited".¹⁷

Robotic pets work in the same way in care homes. PARO, a robotic seal, is listed by the US Food and Drug Administration as a biofeedback device exempt from premarket submission, not as a cleared device.²⁶ In a publicly funded Australian trial in 28 care homes with 415 residents with dementia, PARO kept residents more engaged than a look-alike plush toy, but on the standard agitation scale there were no significant differences between the groups, and a costing found the plush toy "offered marginally greater value for money".¹⁶

Watching for falls

The care-home AI with the clearest operator results watches residents rather than touching them. Cogir Senior Living expanded SafelyYou's AI video fall detection to 36 communities in 2024, reporting "a 52% reduction in the number of residents falling" for residents on the system at least 60 days, comparing their first month with their second.⁹ In England, four Surrey care providers piloted Ally Cares' acoustic monitoring, which uses sound and motion sensors and AI to check residents alone in their rooms at night. The local NHS board reported "a 50 per cent reduction in hospital admissions", "a 57 per cent reduction in total falls" and "over five carer hours per night" saved, and funded 11 more homes.¹⁰ Both are before-and-after figures from the operators and commissioners, not controlled trials. In each case the AI raises the alert and a carer goes to the resident.

Singapore and China: delivery, bathing and exercise

Singapore's Ministry of Health told Parliament in 2023 that its productivity grant had helped 16 nursing homes deploy autonomous mobile robots to deliver "food, linen and other consumables", and that "the use of AMRs has led to at least 10% savings in staff time, allowing them to dedicate more time to direct care provision".¹² At The Salvation Army Peacehaven, a humanoid called Dexie has led pre-programmed exercise, bingo and sing-along sessions for residents with dementia for about two years. One resident told Rest of World: "I don't care for the robot … Makes no difference to me," though he liked playing bingo with it.¹⁸

In China, Xinhua visited the Wuzhen Smart Elderly Care Centre, where meal-delivery robots take boxes to bed-bound residents' doors and a robot leads exercise. The centre's director said bathing a bed-bound resident used to need two people working together, and that with a portable bathing machine "one person can operate it" without moving the resident from bed. The same report noted a Shenzhen care home where a lifting robot "made the elderly clearly uncomfortable" because its centre of gravity didn't suit the human body, and homes where robots that couldn't understand dialect went unused.¹¹ In Chengdu, Leju Robotics' Kuavo humanoid leads tai chi and tells stories at a care home that Xinhua didn't name.²⁰

China's industry and civil-affairs ministries launched a national pilot for 2025 to 2027. Each product must be tested for at least six months, in at least 200 households for home robots or 20 communities or care homes for the rest.²³ The September 2025 list has 32 projects across 10 scenarios, including companionship, rehabilitation, transfer and assisted walking, from companies such as UBTech, Leju, JD and Haier Robotics.²² We found no published results by October 2026.

Toileting and remote help

Seoul is trialling two excretion-care robots at the Seoul Municipal Nursing Home that "automatically handle bowel movements and washing for bedridden patients", with plans to extend them to 11 homes after safety and effectiveness checks, along with 10 wearable robots that assist the wearer's muscles with up to 15 kg.⁸ In Glasgow, Blackwood Homes and Care used a Temi robot as a "first responder" across three buildings: it travels to a resident's flat on its own during emergencies, then "the responder from the remote care team takes control of the robot, assesses the situation and can speak to the resident".¹⁹ That's teleoperation by a person, not the robot giving care.

What the evidence does and doesn't show

Measured against the role's core tasks, AI and robots do two in production: reminding people to take medicine and holding conversations, both through companion robots in older people's homes. They help carers with five more, through sensors that flag who needs attention, aids that make lifting and bathing a one-person job, and robots that carry trays. The highest-rated task, turning a bed-bound person, has only a lab prototype, and we found no robot that feeds, dresses or washes a resident in real care work. Measured benefits are mostly staff time and falls, reported by operators and governments before and after, not by controlled trials; the trials of companion robots show small or short-lived effects on mood.

No operator here has linked a care robot to job cuts. The sources frame them as a response to shortages: Xinhua cites a report that China will be short of more than 5 million eldercare workers over the next five years,¹¹ and in Japan's survey the reason most often given for not adopting any of these technologies was cost.² Those are the stated contexts, not measured effects of the robots.