Skip to content
problem genome

One of 592 unsolved problems in the Problem Genome, a collection written up for student design and engineering teams.

health · digital

a spoon that comes up empty

Table-top feeding robots for people who cannot lift their arms still miss bites and need pre-cut food, and in US studies a research robot took 1 to 2.4 minutes per bite on campus and a home user's carers said they hand-fed when rushed

Source: Nanavati et al., “Lessons Learned from Designing and Evaluating a…” Open ↗source: primaryLast updated: 2026-11-17

Problem statement

People with severe impairment of both arms, from spinal cord injury, cerebral palsy, ALS, muscular dystrophy and other conditions, often cannot bring food to their own mouth, so someone feeds them at every meal. A man with quadriplegia who worked with one robotics team said: "For a long time, I would only let my mom feed me… If I can have a robot do it… it would be me feeding me, and that would be huge." Table-top feeding robots have been sold for decades. The related-work section of a 2025 HRI conference paper says that of the commercial systems of the last two decades, "all but Obi and Neater Eater were discontinued," and suggests this "may be due to" fixed scooping and delivery motions that led to "acquisition failures, dropped food, or neck strain." (Another device, Mealtime Partner, is still on sale per the Obi3 study.) A 2025 systematic review likewise says adoption "remains limited" and that handling diverse food types is unresolved. The current generation still shows the same weak points. In a one-week home trial of the Obi3 robot, sponsored by its maker, users and carers rated usability highly. But one user and one carer reported the spoon frequently coming up empty, and others found the device too big for a wheelchair tray or a restaurant table. A research system with cameras and force sensing, used by six people with motor impairments on campus and by one of them at home for five days, took 1 to 2.4 minutes per bite in the campus meals (each user's median), against 18 to 30 seconds for people without disabilities. It could not handle some foods, and his carers said that when he was rushed "we don't have time to set [the robot] up. So we have to feed him." (In the study itself, researchers did the setup and cut the food.) The unsolved problem is a feeding device that reliably picks up ordinary food, sets up in less time than hand-feeding, and fits where people actually eat.

Why this matters

Being fed by others can make people feel "self-conscious, pressured, or burdensome," in the HRI paper's words, and in the Obi3 study carers reported moderate to high stress, especially balancing care with work. One parent said a robot could be a "game changer" after "hand-feeding our child three meals a day for ten years." The benefits are real when the device works. In the Obi3 trial, self-feeding impairment ratings fell in all 15 users with arm impairments, from an average of 3.80 to 0.60 on a 0–4 scale. The man in the home study rose from "dependent" to "supervision" at meals. But adoption depends on the everyday failures: a missed scoop, a slow bite, a device that will not fit on the table.

What’s been tried and why it hasn’t worked

(1) Fixed-motion table robots (Bestic, Obi, Neater Eater and others). These improved users' well-being, per the HRI paper's related-work section, but all except Obi and Neater Eater were discontinued. The authors suggest this "may be due to" fixed bite acquisition and transfer motions. (2) Obi3 (2026 study, 50 enrolled, 42 completed; 15 users aged 5 and up, 14 carers, 13 providers; one week at home). Usability scores averaged 82.8 to 85.2 on the System Usability Scale, there were no adverse events, and carers reported lower workload. In interviews, though, one user and one carer raised "the spoon frequently coming up empty" (the authors' summary), and an adult user said "the spoon couldn't scoop food from the edge." One adult user likened the device to a "briefcase" that takes up half the table, which they felt detracted from the social side of eating out. A carer said it was "a little too big for the wheelchair tray." The authors answer a carer's request for a fork or knife by calling it "a potential misunderstanding" of intended use: food must be pre-cut into bite-sized pieces. In our reading, this puts preparation work back on the carer. The study was sponsored by the manufacturer, ran one week, and used a self-feeding rating scale that has not been formally validated. (3) A perception-based research robot (2025). A Kinova arm with a camera and force-sensing fork was built with two community researchers with quadriplegia. For all six users it neared or passed the 80% bite-pickup rate that prior work called practical on their most successful food (and for all but two across the whole meal); transfer success was 94–100%. Users rated robot feeding above carer feeding for sense of control and independence. The system's design goal still assumes pre-cut, bite-sized food. But in the campus meals a bite took 1:00 to 2:26 minutes (each user's median). In the five-day home study, autonomous pickup reached at least 80% in only 5 of 10 meals, and the other meals mixed autonomy with the user steering the arm directly through the web app. The robot could not handle some foods, such as ramen. Carers said, "We need to choose foods that CR2 likes and the robot can pick up," and, "Sometimes, CR2's in rush. So we don't have time to set [the robot] up. So we have to feed him." The authors list missed bites, the limited range of foods, and an approach at eye level that no customisation could fix as open limitations, and suggest co-designing setup and maintenance with carers. (4) A utensil-side fix: Kiri-Spoon (2025 preprint). A soft kirigami spoon, closed by a small linear actuator wired into the robot, was fitted to an Obi in place of its standard spoon and tried by 4 residents with upper-limb impairments at a residential community for adults with physical disabilities. Success rates were similar to the standard spoon: both picked up cereal every time, and Kiri-Spoon was "marginally better" on canned oranges and macaroni and cheese. Half preferred each. The authors say N = 4 was too small for statistical tests. In the preprint's own bench tests on a research robot arm, with no users, its builders report that Kiri-Spoon "outperforms the traditional utensils in acquiring slippery foods like jello and tofu" but "struggles to pick flat foods like lettuce." In a third study in the same preprint, sixteen users without disabilities needed fewer attempts and picked up more food with Kiri-Spoon than with a standard fork and spoon, but "found it more comfortable to eat morsels from traditional utensils". So a utensil change has been tried; it adds an actuator and power, and with the four users who have arm impairments it did not clearly beat the plain spoon.

What would unlock progress

In our reading, the two studies point to the same weak link: picking up the bite, and the setup around it. The robot arm itself is not the main problem. Between them, the studies on this page name concrete gaps: scooping from the edge of a bowl, food that is soft, small or slippery, a utensil and dish pairing that turns a miss into a hit, a smaller footprint for a wheelchair tray or a café table, and setup in under the time a carer would take to just do it. These are partly utensil, dish and food-preparation design, not only robot control, so progress does not have to wait for a better robot. The one utensil fix found (Kiri-Spoon) is actuated and showed only marginal gains with four users; we found no study that changed the bowl or dish side for a fixed scooping path. A passive dish or utensil change, needing no power, is the open space.

How we’d know it’s being solved

(a) Bite pickup success per food type, with the 80% mark from prior work as the floor, measured on the foods users actually choose. (b) Time per bite against the 18 to 30 seconds of unassisted eating and the 1 to 2.4 minutes of the research robot. (c) Setup and clean-up time compared with the time a carer takes to hand-feed the same meal. (d) The share of meals over several weeks for which the device is used rather than skipped. (e) The footprint on a standard wheelchair tray and a café table.

Where to start

Facility-free door (make a thing): design a dish and utensil pair that raises scoop success for a fixed, repeated scooping motion, for example a bowl wall profile that lets a spoon scoop cleanly from the edge, or a utensil edge for soft and small foods. Test it with a simple hand-cranked or servo-driven jig that repeats one scoop path on a set of foods. Count hits, misses and spills against a standard bowl and spoon. Use the foods Kiri-Spoon tested (cereal, canned oranges, macaroni and cheese) so the food set matches the one earlier utensil test; that test ran on a robot with users, so jig counts are a bench comparison, not a like-for-like match. Keep the design passive (no actuator). No robot, partner or restricted data is needed: an ordinary workshop, a 3D printer and supermarket food are enough. Reachable users door: people who cannot lift their arms, from high spinal cord injury or other severe arm impairment, and the carers who feed them. A team can reach them within a term through a campus disability office, a local disability peer group or a local independent-living centre. In short sessions they handle the dish and utensil prototypes, walk through their carers' food-preparation steps, and map where a feeding device would have to fit on their own trays and tables. No one is fed in these sessions. Ethics review by the campus board is needed before any session with them. Design-the-test door: write a multi-week home-use protocol measuring (a)–(e), including meals the device was skipped for and why. It is handed to a team that owns a feeding robot, not run. Relevant skills: product and mechanical design, food science, occupational therapy input, human-robot interaction.

Genes

Related problems

Stuck the same way, in other fields.

field: food & the living worldninety-eight percent of u.s. chemicals, never assessedOnly 1.8% of U.S. commercial chemicals have a published human health assessment from EPA or ATSDRtested in the wrong worldsource: primaryfield: digital & spacewho made thisSynthetic content provenance verification gaptested in the wrong worldsource: primaryfield: digital & spaceno eyes on the cold universe since 2013Far-infrared astronomy is blocked by detector arrays too small and not sensitive enough for a space missiontested in the wrong worldsource: primaryfield: food & the living worldmining the abyss, blind to the plumeDeep-sea mining sediment plume monitoring and predictiontested in the wrong worldsource: secondary

Sources

  • Nanavati A, Gordon EK, Faulkner TAK, Song YR, Ko J, Schrenk T, Nguyen V, Zhu BH, Bolotski H, Kashyap A, Kutty S, Karim R, Rainbolt L, Scalise R, Song H, Qu R, Cakmak M, Srinivasa SS. "Lessons Learned from Designing and Evaluating a Robot-assisted Feeding System for Out-of-lab Use." ACM/IEEE International Conference on Human-Robot Interaction 2025:696-707, doi:10.1109/hri61500.2025.10974182, accessed 2026-11-17. Link ↗
  • Fairman AD, Osal RB, Keeton H, Dekar JP. "Usability and Functional Outcomes of the Obi3 Robotic Feeding Device Among People With Disabilities, Caregivers, and Providers: Mixed Methods Study." JMIR Rehabilitation and Assistive Technologies 2026;13:e86338, doi:10.2196/86338, accessed 2026-11-17. Link ↗
  • Keely M, Franco B, Grothoff C, Jenamani RK, Bhattacharjee T, Losey DP, Nemlekar H. "Kiri-Spoon: A Kirigami Utensil for Robot-Assisted Feeding." arXiv:2501.01323, 2025 (preprint), accessed 2026-11-17. Link ↗
  • Liu F, Li Z, Hu M. "Robot-assisted feeding: A systematic review and future prospects." Technol Health Care 2025;33(5):2320-2341, doi:10.1177/09287329251342392 (full text read through library access), accessed 2026-11-17. Link ↗
Revision notes

This brief gets checked against its sources and contains the following revisions. Future revisions get added here.

Full check

Sources read

  • Nanavati et al., feeding robot studies on campus and at home (HRI, 2025): full text
  • Fairman et al., Obi feeding robot home trial (JMIR Rehabilitation and Assistive Technologies, 2026): full text (sponsored by the robot's maker)
  • Keely et al., Kiri-Spoon utensil (preprint, 2025): full text
  • Liu, Li & Hu, systematic review of robot-assisted feeding (Technology and Health Care, 2025): full text
  • Jung, Jeong & Kwon, care robot study (Medicine, 2024): full text

What changed

  • Not found in the sources, replaced with what they say: the place where Kiri-Spoon was tried is now a residential community for adults with physical disabilities, as the preprint says.
  • Stated more strongly than the source: that Kiri-Spoon did not clearly beat the plain spoon is now limited to the four-user test; a sentence now gives its bench result, better on slippery foods and worse on flat ones.
  • Contradicted by the source, corrected: the 1 to 2.4 minutes per bite comes from the campus meals and is each user's median; the paper gives no bite time for the home study, and the page now says the home user was one of the six.
  • Numbers: the fall in impairment scores from 3.80 to 0.60 is now given as a group average.
  • Title: the long title now places the per-bite time on campus and the hand-feeding remark with a home user's carers (display title unchanged).

Left open

  • that most fixed-motion feeding robots were discontinued (the page gives the authors' hedged sentence; the papers behind it were not opened)
  • that people who cannot lift their arms are fed by someone at every meal (no source sentence)

Separate check of specific claims: A larger test of the Kiri-Spoon in users without disabilities was added, and the preprint and the review were added to the source list.

Earlier full check: The discontinuation claim was brought back to the paper's hedged wording, how the home user steered the arm was corrected, which study names which gap was spelled out, and two notes on other studies were corrected.

Separate check of specific claims: The make starting point now uses the same foods as the earlier utensil test and says what it takes, and the users starting point names carers, rules out feeding anyone, and requires ethics review first.

Earlier full check: The long title was toned down (the discontinuation statement rests on one hedged sentence, and the rushed-carer remark on one carer), several statements were narrowed to the users who made them, and an earlier utensil test and a 2025 systematic review were added.