FEAST: A Flexible Mealtime Assistance System
Towards In-the-Wild Personalization

🏆 Best Paper Award, RSS 2025 🏆
FEAST enables care recipients to personalize mealtime assistance in-the-wild, with minimal researcher intervention across diverse in-home scenarios.
Benjamin having dinner in a personal context
10x
Aimee having dinner in a social context
10x
Benjamin having dinner while watching TV
10x
Aimee having dinner in a personal context
10x
Benjamin having lunch in a social context
10x
Aimee having dinner while watching TV
10x

Technical Summary Video

Shaped by Community-Based Participatory Research and a Formative Study

FEAST has been built with Aimee (who has Multiple Sclerosis) and Benjamin (who has a Spinal Cord Injury), through ongoing collaboration since Nov 2022. We begin with a formative study using speculative videos to spark in-depth conversations with 21 care recipients. This highlights three key tenets for mealtime assistance: adaptability, transparency, and safety, and reveals 46 distinct personalization needs (details in the paper).

Placeholder image

Designed for Personalization in Both Hardware and Software

FEAST tackled the identified requirements through: (i) modular hardware that enables switching between feeding, drinking, and mouth-wiping, (ii) diverse interaction methods, including a web interface, head gestures, and physical buttons, to accommodate diverse functional abilities and preferences and (iii) parameterized behavior trees that can be safely and transparently adapted using a large language model.

Placeholder image

Tackles the Three Tenets of In-the-Wild Personalization

For adaptability, we show FEAST meets 36 / 46 adaptability requirements (significantly more than baselines). For Transparency, it tackles Levels 1 to 5 of the IEEE Transparency Standard which includes answering user-initiated queries and providing continuous explanations of what it is doing. For safety, FEAST is designed in accordance with an ISO standard for Care Robots.

Placeholder image

Users Feel MORE Independent Than With Human Caregivers

In a five-day in-home study, care recipients completed meals with minimal researcher intervention, successfully personalized FEAST to their needs, and reported low cognitive workload (NASA-TLX). They rated the system highly for real-world use (Technology Acceptance Model) and felt it gave them greater control and independence than their human caregiver.

Placeholder image

Lessons Learned from Five-Day In-Home Study

We distill five key lessons for deploying assistive robots in real homes. Click each lesson to view an example video with sound.

Placeholder image

Meet the Team

This work would not be possible without the awesome team.
Special thanks to Benjamin and Aimee for shaping this work and welcoming us into their homes.

Rajat Kumar Jenamani
Cornell University
Tom Silver
Cornell University
Ben Dodson
Cornell University
Shiqin Tong
Cornell University
Anthony Song
Cornell University
Yuting Yang
University of Michigan
Ziang Liu
Cornell University
Benjamin Howe
Independent Researcher
Aimee Whitneck
Independent Researcher
Tapomayukh Bhattacharjee
Cornell University

Citation

@inproceedings{jenamani2025feast,
  title     = {FEAST: A Flexible Mealtime-Assistance System Towards In-the-Wild Personalization},
  author    = {Rajat Kumar Jenamani and Tom Silver and Ben Dodson and Shiqin Tong and Anthony Song and 
               Yuting Yang and Ziang Liu and Benjamin Howe and Aimee Whitneck and Tapomayukh Bhattacharjee},
  booktitle = {Robotics: Science and Systems (RSS)},
  year      = {2025},
}