Kitchen Companion for Healthy Ageing

The Challenge

Loneliness and malnutrition — two sides of the same crisis

In Sweden, 20% of the population is aged 65 or older — a share projected to reach 23% by 2040. Among those aged 75 and older, nearly half live alone. This is not simply a demographic fact: it is a daily reality that erodes both nutrition and social connection in a mutually reinforcing cycle.

Loneliness is a serious public health issue linked to cognitive decline, poor mental and physical health, and increased mortality. The absence of social companionship decreases the motivation to cook fresh, nutritious meals, leading to meal skipping and over-reliance on nutrient-poor ready-made food. The kitchen — traditionally a hub of nourishment and social activity — becomes a site of anxiety and neglect.

Kitchen Companion re-engages older adults in meaningful cooking through a proactive, voice-based AI assistant. The system is designed to empower, not replace, user agency: it guides, encourages, and converses throughout the cooking process, rebuilding confidence and restoring the kitchen as a place of daily purpose and social meaning.

Building on Food Talk (Prata Mat)

Kitchen Companion extends the Vinnova-funded Food Talk / Prata Mat project carried out by KTH, Electrolux, and Nagoon in the KTH Intelligence Augmentation Lab. That project validated voice-guided cooking assistance with older adults and surfaced key design requirements — including proactive engagement, pacing adaptability, and privacy-conscious interaction — that now form the foundation of this work.

47%

Living alone at 75+

Nearly half of Swedes aged 75 and older live in single-person households, one of the world's highest rates.

1.8×

Higher malnutrition risk

Older adults living alone face a risk of malnutrition 1.8 times higher than those who cohabitate.

50%

Report loneliness

Between one-third and one-half of elderly people in Sweden report feeling lonely at least sometimes.

36k+

Mortality link confirmed

A Swedish cohort of over 36,000 participants confirmed high social isolation significantly increases all-cause mortality risk.

Previous Work

Food Talk: A Voice Assistant for the Kitchen

Kitchen Companion builds directly on Food Talk (Prata Mat), a Vinnova-funded project carried out by KTH, Nagoon, and Electrolux. The project developed and evaluated a Wizard-of-Oz voice assistant for elderly home cooks — providing the empirical foundation and key design lessons for the current work.

The study

Six elderly Swedish participants (ages 63–66, five living alone) cooked a mushroom omelette in the KTH Intelligence Augmentation Lab, guided by a voice assistant named “Bengt”. Two system versions were compared: an Instructional AI Chef that gave step-by-step cooking instructions only, and a Chatty AI Chef that supplemented instructions with encouragement, personal questions, and food-related trivia.

A human operator monitored live video feeds and controlled the assistant in real time — deciding when the user had completed each step and when to interject with social content. All prompts were pre-rendered using a Swedish text-to-speech engine and delivered via speakers in the lab kitchen.

Prior to the main study, two pilots were conducted: one comparing LLM-based cooking step detection (Gemini 2.5 Flash vs. LLaVA) from multiple camera angles, and one exploring personalisation across four user profiles — from Beginner to Creative Expert — each with distinct instruction preferences.

Key finding

Chattiness served functional purposes beyond entertainment. The instructional group interpreted natural cooking pauses as slowness and system unawareness — “dead air”. The chatty group experienced the same pauses as engaging, and perceived the assistant as more intelligent and responsive. Social dialogue transformed waiting time into collaborative experience rather than simply speeding up instructions.

“I thought that was really fun — while you were still waiting for something to be ready, you could hear some stories.”

Chatty AI Chef participant

Results at a glance

6

Participants

Ages 63–66 · 5 living alone · 2 men, 4 women

2

System versions

Instructional only vs. Chatty (instructions + trivia + encouragement)

3/3

Chatty group satisfied

All chatty participants perceived the assistant as smart and action-aware

2/3

Instructional felt slow

Two of three found the session “long and slow”, perceiving the assistant as a simple timer

Lessons carried forward

→ Proactive social dialogue is not a luxury — it is what makes a voice assistant feel situationally aware
→ Current LLMs cannot yet detect cooking step completion reliably from video in real time — motivating the architectural choices in Kitchen Companion
→ User profiles (Beginner to Creative Expert) require distinct instruction styles — personalisation is essential, not optional
→ Privacy attitudes are divided — transparent, edge-only processing is a prerequisite for home deployment
→ All project goals were met; the validated prototype is ready for the next stage of development

Publication

Esteve, Fredriksson, Gustafson, Kontogiorgos & Mashiyi-Veikkola. Towards a proactive cooking companion for the elderly. IWSDS 2026.

Industry Partner · LinkedIn

Electrolux Group

View post →

“Bengt brings a sense of presence, encouragement, and shared purpose to everyday cooking.” As PI of the Vinnova-funded Prata Mat project, Electrolux led user studies conducted in the KTH Intelligence Augmentation Lab together with KTH and Nagoon — investigating how conversational AI can support recipe guidance, meal planning, and independent living for older adults.

108 reactions · Electrolux Group (759K followers)

Research Questions

What we set out to answer

Three research questions, defined in collaboration with public-sector partners and refined through prior pilot work, guide the project's scientific agenda.

RQ 1 — Contextual Barriers

Situated barriers in the kitchen

How do the specific "situated" barriers of the kitchen environment — from physical dexterity challenges to the cognitive load of multi-tasking while cooking — interact with social isolation to inhibit healthy cooking among older adults?

RQ 2 — System Design

Multimodal grounding & proactivity

How can a conversational AI architecture be designed to leverage multimodal grounding (fusing vision and audio) to transition from reactive command-execution to proactive, context-aware assistance without infringing on user autonomy?

RQ 3 — Longitudinal Impact

Restored capability & social eating

To what extent can a proactively interactive Kitchen Companion improve cooking self-efficacy and nutritional habits — and does restored confidence lead to increased social eating behaviors, such as preparing meals for others or sharing meals, over time?

Method & Implementation

Three phases, older adults at every step

The project follows a proven participatory design framework adapted from the Age-Wear project, also funded by the Kamprad Family Foundation, treating older adults as co-designers throughout the full 36-month lifecycle.

Phase 1 · Months 1–12

🔍

Understand

Deep contextual inquiry to build a rigorous understanding of older adults' cooking routines, nutritional challenges, and social needs before any design decisions are made.

Survey study (n=200, Stockholm region) using UCLA Loneliness Scale, SCREEN-II nutritional risk assessment, and technology literacy scales
In-depth elicitation interviews in participants' own kitchens
Low-fidelity technology probes to surface unmet needs and implicit technology norms
Multimodal sensor specification for typical elderly kitchens

Led by Sanna Kuoppamäki & Malin Ljung

Phase 2 · Months 10–24

🛠

Co-Create

Hands-on design workshops where older adults function as designers, not merely consumers — shaping the assistant's voice, personality, conversational style, and error-handling strategies.

Wizard-of-Oz cooking sessions: participants cook real meals while a human operator simulates the AI
Role-playing future scenarios, dialogue scripting, voice and persona design exercises (~30 participants)
Parallel technical development: Context-aware Interaction Manager, LLM finetuning, visual grounding pipeline
Design of interaction modes — Encouraging Mode and Efficient Mode — with matching voice and tone

Led by Joakim Gustafson & Sanna Kuoppamäki

Phase 3 · Months 25–36

✅

Validate

Rigorous assessment of the Kitchen Companion's real-world impact through controlled in-lab trials and a longitudinal home study.

Structured in-lab evaluation in the KTH IA Lab: task completion, error rate, system acceptance (Godspeed Questionnaire)
2-week longitudinal home study with 8 older adults using a tablet-based prototype
Pre/post surveys on loneliness (UCLA Scale), cooking self-efficacy (EXSE), and shared meal frequency
Culminating cook-along event: participants cook together and share experiences at the KTH IA Lab

Led by Morgan Fredriksson & Joakim Gustafson

A voice-first, context-aware system

The Kitchen Companion is a speech-output only system: it speaks. It listens through speech and audio, and watches through a privacy-preserving camera to understand what is happening in the kitchen — without ever transmitting raw video.

A Context-aware Interaction Manager acts as the system's conductor, coordinating a task thread (step-by-step instructions) and an engagement thread (encouragement, acknowledgment, reassurance). A finetuned local LLM ensures all advice is grounded, safe, and hallucination-resistant.

Input

Speech

Natural spoken language from the user — questions, confirmations, requests

Input

Audio & Vision

Kitchen sounds and locally-processed video for activity recognition — sizzling, chopping, posture, ingredient state

Output

Speech

Natural, adaptive spoken responses — instructions, encouragement, conversational engagement

Privacy

Edge-only

All video processed locally. No raw video transmitted. Visual indicator when the system is active.

Team

An interdisciplinary consortium

The project brings together world-class expertise in conversational AI, gerontechnology, and interaction design — spanning academic research and industry product development.

👤

Principal Investigator

Joakim Gustafson

Professor & Head of Department · KTH Speech, Music and Hearing (TMH)

Leading expert in conversational AI and speech technology. Head of the Department of Speech, Music and Hearing at KTH. His work spans multimodal dialogue systems, social robotics, and speech synthesis — from early spoken dialogue systems in the 1990s through to today's LLM-powered interaction. PI of Food Talk / Prata Mat and of the SAInt humanoid robot project. Editorial board member of Speech Communication and Review Editor at Frontiers in Robotics and AI; Treasurer of ISCA. Provides the KTH Intelligence Augmentation Lab facility.

KTH Profile Research Page

👤

Co-Principal Investigator

Sanna Kuoppamäki

Assistant Professor (tenure track) · KTH Biomedical Engineering & Health Systems

Specialises in user-centric design, gerontechnology, and participatory design with older adults. Leads WP1 and WP2, bringing the methodology developed in the Kamprad-funded Age-Wear project to ensure the "age-inclusive" focus is maintained throughout. Her background in sociology ensures a deep understanding of the social implications of the technology. 2024 CHI Best Paper for "Cooking with agents: Designing context-aware voice interaction." Co-PI on NordForsk's Nordic Model for assistive care robots (2025).

KTH Profile

👤

Industry Co-PI · Nagoon AB

Morgan Fredriksson

CEO, Nagoon AB

Leads WP3 (Validate) and the translation of research prototypes into deployable products. Co-founded Nagoon AB (2015), which builds real-time spatial computing and AI voice interfaces. Co-PI on the Vinnova-funded Food Talk / Prata Mat project (2024–2025). Extensive history in accessible technology for vulnerable groups — including OLA, Saapho (AAL), and EmpowerME (PTS Innovation winner). Nagoon's Spatial Platform provides the "digital twin" of the kitchen used for home deployment.

Nagoon

Supporting Partners

🏙

Public Sector Partner

Malin Ljung · Stockholm City

Head of the Healthcare and Development Unit, managing the Verklighetslabbet (Reality Lab) partnership. Facilitates access to testbeds and supports participant recruitment across all work packages and dissemination to municipal elderly care.

🏠

External Advisor

Katarina Esteve · Electrolux

Head of the Global Accessibility Team, Electrolux Design Community. Was PI of the Vinnova-funded Food Talk project with KTH and Nagoon. Serves as external advisor, bridging research findings with consumer appliance design perspectives.

Infrastructure

KTH Intelligence Augmentation Lab

All participatory design and validation studies are conducted in the KTH Intelligence Augmentation (IA) Lab — the same facility used in the Vinnova-funded Food Talk project with Electrolux and Nagoon. The lab provides a home-like smart kitchen environment with synchronized sensors for multimodal activity detection and a separate control room for human-in-the-loop Wizard-of-Oz sessions.

The facility's ecological validity — a real, fully functioning kitchen rather than a sterile lab — is crucial for developing a system that will work reliably in users' own homes. Its GPU servers and real-time video infrastructure support the computationally intensive activity recognition and LLM inference required by the system.

Part of KTH IRL

The Intelligence Augmentation Lab is one of six purpose-built spaces within KTH IRL (Interaction and Robotics Lab) — a shared research facility jointly operated by the Departments of Speech, Music & Hearing and Robotics, Perception & Learning, advancing human-robot interaction and AI research through professional-grade infrastructure.

🎙

Multi-microphone Array

Captures speech in realistic noisy kitchen conditions — sizzling, water running, clattering utensils.

📷

Multi-camera Setup

Overhead and counter-level cameras for real-time action recognition, pose estimation, and ingredient state detection.

🖥

Wizard-of-Oz Console

Control room with GPU servers and live feeds, enabling human operators to simulate the AI during early-stage co-design sessions.

🏠

Home-like Environment

A fully functioning kitchen that mirrors real domestic settings, ensuring findings transfer to participants' own homes.

KTH Intelligence Augmentation Lab — smart kitchen

Sister Project

SAInt — Robotic Partners for Life

SAInt is a closely related research project led by the same PI, Prof. Joakim Gustafson, and shares Kitchen Companion’s core motivation: supporting elderly people in maintaining independent and meaningful lives at home. Where Kitchen Companion uses a voice-only AI assistant, SAInt takes the embodied route — deploying full humanoid robots as domestic partners.

Both projects work from the same smart kitchen environment at KTH and address overlapping research challenges around proactive, situation-aware dialogue. Together they explore the spectrum from purely voice-based to physically embodied AI assistance for healthy ageing. SAInt is funded by Stiftelsen Promobilia.

Visit the SAInt project →

Kitchen Companion

Voice-based AI

Speech output only · multimodal sensing · co-designed with older adults · home deployment focus

SAInt

Humanoid Robots

Embodied physical presence · manipulation · full human–robot interaction · domestic assistance

Kitchen Companion for Healthy Ageing

Loneliness and malnutrition — two sides of the same crisis

Living alone at 75+

Higher malnutrition risk

Report loneliness

Mortality link confirmed

Food Talk: A Voice Assistant for the Kitchen

The study

Key finding

Results at a glance

Participants

System versions

Chatty group satisfied

Instructional felt slow

Lessons carried forward

What we set out to answer

Situated barriers in the kitchen

Multimodal grounding & proactivity

Restored capability & social eating

Three phases, older adults at every step

Understand

Co-Create

Validate

A voice-first, context-aware system

Speech

Audio & Vision

Speech

Edge-only

An interdisciplinary consortium

Joakim Gustafson

Sanna Kuoppamäki

Morgan Fredriksson

Supporting Partners

Malin Ljung · Stockholm City

Katarina Esteve · Electrolux

KTH Intelligence Augmentation Lab

Multi-microphone Array

Multi-camera Setup

Wizard-of-Oz Console

Home-like Environment

SAInt — Robotic Partners for Life

Join the project

Elderly Care Organisations

Appliance & Care Technology Companies

Questions about the project?

Kitchen
Companion
for Healthy Ageing