The intricacies of eating are more intricate than they may seem. For nearly ten years, scientists at the Personal Robotics Lab of the University of Washington have been endeavoring to develop a robot that can assist individuals unable to feed themselves.
The initial breakthrough for the researchers, conducted when the lab was associated with Carnegie Mellon University, was enabling a robotic arm to utilize a fork to feed a person a marshmallow. Since then, the robot has evolved from feeding users fruit salads to full meals consisting of nearly any food that can be picked up with a fork. The researchers have also explored ways in which the robot can enhance the social aspects of dining.
Until recently, this research primarily took place in controlled laboratory settings. However, last year, researchers used the assistive feeding arm in two studies outside the lab. The first study involved six participants with motor impairments who utilized the robot to feed themselves in a UW cafeteria, an office, or a conference room. In the second study, one participant, Jonathan Ko, a community researcher and co-author of the study, employed the system at home for five days, where it fed him ten meals.
The research team will present their findings on March 5 at the ACM/IEEE International Conference on Human-Robot Interaction in Melbourne.
“In our previous studies, we operated in laboratory settings because controlling all the meal aspects is essential for evaluating specific system components in isolation,” explained lead author Amal Nanavati, a doctoral student at the Paul G. Allen School of Computer Science & Engineering at UW. “However, this approach fails to encompass the various meal contexts encountered outside of the lab. Ultimately, our objective is to empower individuals to feed themselves in authentic environments, so it is vital to assess the system within those contexts.”
The device, referred to by researchers as ADA (Assistive Dexterous Arm), comprises a robotic arm that can be attached to adjacent objects, such as a powered wheelchair or a hospital table. The user requests a specific bite through a web application, and the system then autonomously feeds the user that bite (though users have the option to halt the arm using a “kill button”). The arm is equipped with a force sensor and a camera to differentiate between various foods and deliver the food to the user’s mouth.
In both studies, participants successfully fed themselves their meals. In the first study, the robot achieved around 80% accuracy in procuring entrees, a threshold deemed successful by users in another study. During the second study, the diverse circumstances and surroundings at home — such as Ko possibly eating while watching TV in dim light or working from bed — posed challenges to the system’s default functionality. However, researchers designed the system to allow for customization, enabling Ko to manage the robot and successfully feed himself all meals.
The team intends to further enhance the system for effectiveness and adaptability.
“Taking the robot outside of the lab was a significant milestone,” Ko remarked. “There are various settings where you eat, and many small variables that can easily be overlooked. If the robot is excessively heavy, it could tilt a table. Alternatively, inadequate lighting might hinder facial recognition, which is something you don’t typically think about while dining.”
Additional co-authors encompass Ethan K. Gordon, Bernie Hao Zhu, and Rosario Scalise, all doctoral candidates in the Allen School; Taylor A. Kessler Faulkner, a lecturer in the Allen School; Tyler Schrenk, the late president of the Tyler Schrenk Foundation and a community researcher; Vy Nguyen, an occupational therapy clinical research lead at Hello Robot; Haya Bolotski, a high school intern in the Allen School during this project; Yuxin (Ray) Song, Atharva Kashyap, Sriram Kutty, Liander Rainbolt, and Ramon Qu, all undergraduate students in the Allen School during this research; Raida Karim, a research scientist assistant in the Allen School during this project; and Hanjun Song, a research scientist in the Allen School while undertaking this research. Maya Cakmak and Siddhartha S. Srinivasa, both professors at the Allen School, serve as senior authors. Funding for this research was partially provided by a UW CREATE student grant; UW Allen School Postdoc Research award; the National Science Foundation; the DARPA RACER program; the National Institute of Biomedical Imaging and Bioengineering; and the Office of Naval Research.
For further information, reach out to Nanavati at [email protected].