WAFFLE predicts bite timing
Researchers presented WAFFLE, a system using wearable sensors and machine learning to predict when a person will bite based on cues like chewing and motion. The work—recognized at HRI 2026—illustrates how fine‑grained sensor fusion can enable new interaction models for feeding assistance and diet‑aware apps. (x.com)
Robotic feeding systems still struggle with one basic question: when should the robot move food to a person’s mouth? WAFFLE is a new wearable system built to estimate that moment from head motion, chewing, and speech cues. (arxiv.org) The system’s full name is “Wearable Approach For Feeding with LEarned Bite Timing,” and the paper was presented at the 21st ACM/IEEE International Conference on Human-Robot Interaction, held March 16–19, 2026, in Edinburgh, Scotland. The conference listed WAFFLE as the Systems Track Best Paper. (dl.acm.org) (humanrobotinteraction.org) WAFFLE uses wearable sensors — including glasses with an inertial measurement unit and a throat microphone — to estimate the time until the next bite. The model was trained on data from 14 participants without impairments and then turned its predictions into “stop” or “progress” commands with a user-set threshold. (dl.acm.org) That timing problem sits at the center of robot-assisted feeding, where a machine has to wait for a user to finish chewing, swallow, and signal readiness without forcing a rigid pace. The WAFFLE paper says poor bite-timing has been one reason feeding robots have not spread more widely. (arxiv.org) The work plugs into a larger push to make feeding robots usable outside labs, especially for people with mobility impairments who want more control over meals. A 2024 HRI companion paper from the same research area said at least 1.8 million people in the United States require assistance to eat. (dl.acm.org) In evaluation, the researchers tested WAFFLE with the Obi feeding robot in a study of 15 participants without motor impairments. The paper reports the system performed statistically on par with or better than baseline methods on perceived control, robot understanding, and workload, and that a majority preferred it in both solo and social dining. (arxiv.org) The team also ran a smaller home study with two participants with motor impairments using a Kinova 7-degree-of-freedom robot arm. The paper says those tests were meant to check whether the method could carry over across users, hardware, foods, robot positions, and meal settings. (dl.acm.org) The author list spans Carnegie Mellon University and Cornell University, with Akhil Padmanabha, Jessie Yuan, and Tanisha S. Mehta among the lead authors, and Tapomayukh Bhattacharjee and Zackory Erickson listed as advising authors. Padmanabha’s public profile says he recently completed a Ph.D. at Carnegie Mellon’s Robotics Institute. (dl.acm.org) (akhilpadmanabha.github.io) What WAFFLE adds is not a new robot arm but a new way to read the diner, using body signals as a timing cue instead of relying only on the robot’s own schedule. In feeding assistance, that can mean the difference between a machine that serves bites on a timer and one that waits for the person. (arxiv.org)
